From f85a712fed1b9e35c8ed687beb7d6717746bee81 Mon Sep 17 00:00:00 2001 From: Yusuke-KOMIYAMA Date: Mon, 16 Nov 2026 12:50:20 +0900 Subject: [PATCH 1/3] =?UTF-8?q?docs(review):=202026-11-12=E3=80=9C16=20?= =?UTF-8?q?=E3=81=AE=E3=83=AC=E3=83=93=E3=83=A5=E3=83=BC=E8=A8=98=E9=8C=B2?= =?UTF-8?q?=2056=20=E4=BB=B6=E3=82=92=E5=8F=8E=E3=82=81=E3=82=8B=EF=BC=88?= =?UTF-8?q?=E3=82=AC=E3=82=A4=E3=83=89=20#57=E3=80=9C#91=E3=80=81cas-overl?= =?UTF-8?q?ay=20#5=E3=80=81aipf=20#1=20=E3=81=BB=E3=81=8B=EF=BC=89?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit このセッションの REV(担当レビュワ, Codex Astra)が scratchpad に書き、まだ docs/reviews/ に無かった 記録を収める。ADR-04 によりレビューは独立した記録として残す。 - RDM-developer-guide: PR #57・#59〜#66・#68・#69・#72・#73・#75・#76・#78・#80・#82〜#89・#91 (#87 の記録は RDM-rdmcp-dev2027 #1、#88 r2 は RDM-cas-overlay-dev2027 #4 も扱う) - RDM-cas-overlay-dev2027 #5 r1〜r4、RDM-aipf_mcp_data_grdm-dev2027 #1 r1〜r3、 RDM-ember-osf-web-dev2027 #4 r1〜r3・#5 r1、RDM-osf.io-dev2027 #22 r1 収める前に、本体のチェックアウトの絶対パスをリポジトリからの相対パスに、scratchpad の絶対パスを に、作業者の利用者名を一般の語に置き換えた(内容と判定は変えていない)。PR #69 の r1 は 別々の REV による二つの記録(11-14 と 11-15)があるので両方を残す。コメントの下書き 1 件は記録でない ので収めない。docs_check は合格。 Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BW1YQk71jLXY2FaSzMnnYJ --- ...ber-osf-web-dev2027_PR4_Zxcvbn_Local_r1.md | 164 ++++++++++++ ...ber-osf-web-dev2027_PR4_Zxcvbn_Local_r2.md | 244 ++++++++++++++++++ ...ber-osf-web-dev2027_PR4_Zxcvbn_Local_r3.md | 124 +++++++++ ...eveloper-guide_PR57_Records_Live_ZAP_r1.md | 88 +++++++ ...eveloper-guide_PR57_Records_Live_ZAP_r2.md | 58 +++++ ...guide_PR59_Ubuntu_Arm64_Cold_Volumes_r1.md | 96 +++++++ ...guide_PR59_Ubuntu_Arm64_Cold_Volumes_r2.md | 58 +++++ ...-guide_PR60_Wait_Scale_Trivy_Timeout_r1.md | 111 ++++++++ ...-guide_PR60_Wait_Scale_Trivy_Timeout_r2.md | 98 +++++++ ...per-guide_PR61_M2_Close_ADR_Approval_r1.md | 122 +++++++++ ...per-guide_PR61_M2_Close_ADR_Approval_r2.md | 56 ++++ ...eveloper-guide_PR62_Amd64_Final_Pins_r1.md | 101 ++++++++ ...eveloper-guide_PR62_Amd64_Final_Pins_r2.md | 75 ++++++ ...loper-guide_PR63_Scrub_Host_Identity_r1.md | 120 +++++++++ ...loper-guide_PR63_Scrub_Host_Identity_r2.md | 97 +++++++ ...loper-guide_PR63_Scrub_Host_Identity_r3.md | 65 +++++ ...loper-guide_PR63_Scrub_Host_Identity_r4.md | 78 ++++++ ...oper-guide_PR64_Install_Volume_Guard_r1.md | 74 ++++++ ...veloper-guide_PR65_Workflow_Scope_CI_r1.md | 173 +++++++++++++ ...veloper-guide_PR65_Workflow_Scope_CI_r2.md | 141 ++++++++++ ...oper-guide_PR66_DEMO22_Uninterrupted_r1.md | 65 +++++ ...eloper-guide_PR68_Status_Publication_r1.md | 69 +++++ ...eloper-guide_PR69_DEMO20_macOS_Fresh_r1.md | 72 ++++++ ...-ember-osf-web-dev2027_PR5_Node24_CI_r1.md | 77 ++++++ ....io-dev2027_PR22_Private_Security_CI_r1.md | 104 ++++++++ ...eloper-guide_PR69_DEMO20_macOS_Fresh_r1.md | 123 +++++++++ ...M-developer-guide_PR72_Exit2_Message_r1.md | 171 ++++++++++++ ...M-developer-guide_PR72_Exit2_Message_r2.md | 119 +++++++++ ...eveloper-guide_PR73_Exit_Table_Check_r1.md | 117 +++++++++ ...eveloper-guide_PR73_Exit_Table_Check_r2.md | 114 ++++++++ ...developer-guide_PR75_osf_Pin_94e6817_r1.md | 138 ++++++++++ ...developer-guide_PR76_Up_Stale_Render_r1.md | 86 ++++++ ...-developer-guide_PR78_S1_y_websocket_r1.md | 88 +++++++ ...-developer-guide_PR78_S1_y_websocket_r2.md | 55 ++++ ...-developer-guide_PR80_rdmcp_Template_r1.md | 129 +++++++++ ...loper-guide_PR82_DEMO22_Empty_Docker_r1.md | 70 +++++ ...15_RDM-developer-guide_PR83_S4_OASys_r1.md | 84 ++++++ ...15_RDM-developer-guide_PR83_S4_OASys_r2.md | 74 ++++++ ...developer-guide_PR84_osf_Pin_07cc56d_r1.md | 123 +++++++++ ...DM-cas-overlay_PR5_NoExternalScripts_r1.md | 95 +++++++ ...DM-cas-overlay_PR5_NoExternalScripts_r2.md | 89 +++++++ ...DM-cas-overlay_PR5_NoExternalScripts_r3.md | 100 +++++++ ...DM-cas-overlay_PR5_NoExternalScripts_r4.md | 86 ++++++ ...M-developer-guide_PR85_Health_Settle_r1.md | 62 +++++ ...M-developer-guide_PR85_Health_Settle_r2.md | 60 +++++ ...DM-developer-guide_PR86_S3_aipf_OIDC_r1.md | 95 +++++++ ...DM-developer-guide_PR86_S3_aipf_OIDC_r2.md | 87 +++++++ ...16_RDM-developer-guide_PR87_S2_rdmcp_r1.md | 114 ++++++++ ...M-developer-guide_PR88_S5_Shibboleth_r1.md | 123 +++++++++ ...M-developer-guide_PR88_S5_Shibboleth_r2.md | 128 +++++++++ ...M-developer-guide_PR88_S5_Shibboleth_r3.md | 66 +++++ ...16_RDM-developer-guide_PR89_aipf_Pin_r1.md | 92 +++++++ ...developer-guide_PR91_osf_Pin_c5a05de_r1.md | 101 ++++++++ .../2026-11-16_aipf_PR1_AccessLog_r1.md | 71 +++++ .../2026-11-16_aipf_PR1_AccessLog_r2.md | 73 ++++++ .../2026-11-16_aipf_PR1_AccessLog_r3.md | 85 ++++++ 56 files changed, 5548 insertions(+) create mode 100644 docs/reviews/2026-11-12_RDM-ember-osf-web-dev2027_PR4_Zxcvbn_Local_r1.md create mode 100644 docs/reviews/2026-11-12_RDM-ember-osf-web-dev2027_PR4_Zxcvbn_Local_r2.md create mode 100644 docs/reviews/2026-11-12_RDM-ember-osf-web-dev2027_PR4_Zxcvbn_Local_r3.md create mode 100644 docs/reviews/2026-11-13_RDM-developer-guide_PR57_Records_Live_ZAP_r1.md create mode 100644 docs/reviews/2026-11-13_RDM-developer-guide_PR57_Records_Live_ZAP_r2.md create mode 100644 docs/reviews/2026-11-13_RDM-developer-guide_PR59_Ubuntu_Arm64_Cold_Volumes_r1.md create mode 100644 docs/reviews/2026-11-13_RDM-developer-guide_PR59_Ubuntu_Arm64_Cold_Volumes_r2.md create mode 100644 docs/reviews/2026-11-13_RDM-developer-guide_PR60_Wait_Scale_Trivy_Timeout_r1.md create mode 100644 docs/reviews/2026-11-13_RDM-developer-guide_PR60_Wait_Scale_Trivy_Timeout_r2.md create mode 100644 docs/reviews/2026-11-13_RDM-developer-guide_PR61_M2_Close_ADR_Approval_r1.md create mode 100644 docs/reviews/2026-11-13_RDM-developer-guide_PR61_M2_Close_ADR_Approval_r2.md create mode 100644 docs/reviews/2026-11-14_RDM-developer-guide_PR62_Amd64_Final_Pins_r1.md create mode 100644 docs/reviews/2026-11-14_RDM-developer-guide_PR62_Amd64_Final_Pins_r2.md create mode 100644 docs/reviews/2026-11-14_RDM-developer-guide_PR63_Scrub_Host_Identity_r1.md create mode 100644 docs/reviews/2026-11-14_RDM-developer-guide_PR63_Scrub_Host_Identity_r2.md create mode 100644 docs/reviews/2026-11-14_RDM-developer-guide_PR63_Scrub_Host_Identity_r3.md create mode 100644 docs/reviews/2026-11-14_RDM-developer-guide_PR63_Scrub_Host_Identity_r4.md create mode 100644 docs/reviews/2026-11-14_RDM-developer-guide_PR64_Install_Volume_Guard_r1.md create mode 100644 docs/reviews/2026-11-14_RDM-developer-guide_PR65_Workflow_Scope_CI_r1.md create mode 100644 docs/reviews/2026-11-14_RDM-developer-guide_PR65_Workflow_Scope_CI_r2.md create mode 100644 docs/reviews/2026-11-14_RDM-developer-guide_PR66_DEMO22_Uninterrupted_r1.md create mode 100644 docs/reviews/2026-11-14_RDM-developer-guide_PR68_Status_Publication_r1.md create mode 100644 docs/reviews/2026-11-14_RDM-developer-guide_PR69_DEMO20_macOS_Fresh_r1.md create mode 100644 docs/reviews/2026-11-14_RDM-ember-osf-web-dev2027_PR5_Node24_CI_r1.md create mode 100644 docs/reviews/2026-11-14_RDM-osf.io-dev2027_PR22_Private_Security_CI_r1.md create mode 100644 docs/reviews/2026-11-15_RDM-developer-guide_PR69_DEMO20_macOS_Fresh_r1.md create mode 100644 docs/reviews/2026-11-15_RDM-developer-guide_PR72_Exit2_Message_r1.md create mode 100644 docs/reviews/2026-11-15_RDM-developer-guide_PR72_Exit2_Message_r2.md create mode 100644 docs/reviews/2026-11-15_RDM-developer-guide_PR73_Exit_Table_Check_r1.md create mode 100644 docs/reviews/2026-11-15_RDM-developer-guide_PR73_Exit_Table_Check_r2.md create mode 100644 docs/reviews/2026-11-15_RDM-developer-guide_PR75_osf_Pin_94e6817_r1.md create mode 100644 docs/reviews/2026-11-15_RDM-developer-guide_PR76_Up_Stale_Render_r1.md create mode 100644 docs/reviews/2026-11-15_RDM-developer-guide_PR78_S1_y_websocket_r1.md create mode 100644 docs/reviews/2026-11-15_RDM-developer-guide_PR78_S1_y_websocket_r2.md create mode 100644 docs/reviews/2026-11-15_RDM-developer-guide_PR80_rdmcp_Template_r1.md create mode 100644 docs/reviews/2026-11-15_RDM-developer-guide_PR82_DEMO22_Empty_Docker_r1.md create mode 100644 docs/reviews/2026-11-15_RDM-developer-guide_PR83_S4_OASys_r1.md create mode 100644 docs/reviews/2026-11-15_RDM-developer-guide_PR83_S4_OASys_r2.md create mode 100644 docs/reviews/2026-11-15_RDM-developer-guide_PR84_osf_Pin_07cc56d_r1.md create mode 100644 docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r1.md create mode 100644 docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r2.md create mode 100644 docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r3.md create mode 100644 docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r4.md create mode 100644 docs/reviews/2026-11-16_RDM-developer-guide_PR85_Health_Settle_r1.md create mode 100644 docs/reviews/2026-11-16_RDM-developer-guide_PR85_Health_Settle_r2.md create mode 100644 docs/reviews/2026-11-16_RDM-developer-guide_PR86_S3_aipf_OIDC_r1.md create mode 100644 docs/reviews/2026-11-16_RDM-developer-guide_PR86_S3_aipf_OIDC_r2.md create mode 100644 docs/reviews/2026-11-16_RDM-developer-guide_PR87_S2_rdmcp_r1.md create mode 100644 docs/reviews/2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r1.md create mode 100644 docs/reviews/2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r2.md create mode 100644 docs/reviews/2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r3.md create mode 100644 docs/reviews/2026-11-16_RDM-developer-guide_PR89_aipf_Pin_r1.md create mode 100644 docs/reviews/2026-11-16_RDM-developer-guide_PR91_osf_Pin_c5a05de_r1.md create mode 100644 docs/reviews/2026-11-16_aipf_PR1_AccessLog_r1.md create mode 100644 docs/reviews/2026-11-16_aipf_PR1_AccessLog_r2.md create mode 100644 docs/reviews/2026-11-16_aipf_PR1_AccessLog_r3.md diff --git a/docs/reviews/2026-11-12_RDM-ember-osf-web-dev2027_PR4_Zxcvbn_Local_r1.md b/docs/reviews/2026-11-12_RDM-ember-osf-web-dev2027_PR4_Zxcvbn_Local_r1.md new file mode 100644 index 00000000..2abc6d58 --- /dev/null +++ b/docs/reviews/2026-11-12_RDM-ember-osf-web-dev2027_PR4_Zxcvbn_Local_r1.md @@ -0,0 +1,164 @@ +# REV PR4 Local zxcvbn review r1 / zxcvbn ローカル配信レビュー r1 + +> **Summary:** Changes requested: the local zxcvbn change is sound on inspected evidence, but the new external-load guard accepts ordinary HTML and JavaScript forms that introduce unlisted CDN loads. +> **概要:** 修正要求。確認できた範囲で zxcvbn のローカル化は妥当だが、新しい外部読み込み検査は、未掲載の CDN 読み込みを追加する通常の HTML・JavaScript 構文を通してしまう。 + +**Jump:** [English](#english) · [日本語](#japanese) + +- Version / 版: 0.1 — 2026-11-12. **Learned:** an exact URL list does not enforce exact loading when its scanner misses valid browser syntax. / URL 一覧が完全一致でも、検査側がブラウザで有効な構文を読み落とすと読み込みを制限できない。 +- Applies to: RCOSDP/RDM-ember-osf-web-dev2027 PR #4, head `3e58dbc580b6c8e7ce617735e86c3bda11a7b6a9`, base `4180cb58bb36855dfc40e0b7262450d08f51d1d6`. +- Reviewer / レビュワ: REV(担当レビュワ, Codex Astra); reviewed / レビュー日: 2026-11-12. +- Related: [Review index / レビュー索引](README.md) · [SPEC-04](../spec/04_acceptance_tests.md) · [SPEC-06](../spec/06_documentation_conventions.md). +- Used by / 利用先: PR #4 revision and re-review / PR #4 の修正・再レビュー、[review index / レビュー索引](README.md). + +--- + + +## English + +[→ 日本語版へ](#japanese) + +### Verdict and scope + +**Changes requested — one P2 finding (R1).** Reviewed all three changed files: `ember-cli-build.js`, `docker/static/postbuild.js`, and `node-tests/docker-static-postbuild-test.js`. Also inspected the lockfile, production build stages, static nginx configuration, renderer, password-strength addon and callers, and existing CI logs. The head was rechecked before reporting. No merge, approval, commit, push, stack operation, or runtime/.env access was performed. + +### R1 — Recognize browser-valid external loads before applying the exact list + +**Location:** `docker/static/postbuild.js:186–205`, especially the call to `attributes(rawAttrs)` at line 199. The helper at lines 98–109 only reads quoted attributes; the new guard reuses it without rejecting unquoted forms or decoding HTML character references. The inline URL expression at line 186 also omits backticks from its allowed preceding delimiters. These are gaps in the newly introduced guard, even though the helper existed before the PR. + +Starting with the passing test fixture, keep the local zxcvbn and both listed URLs, and append each example separately: + +```html + + + +``` + +Each independently exits **0** with `external loads: 2 listed, zxcvbn local`. The equivalent quoted `src="https://cdnjs.cloudflare.com/..."` correctly exits **1**. These constructs can load an unlisted script under the existing `script-src https://cdnjs.cloudflare.com`; the inline loader even receives a generated CSP hash. This is a build-policy regression gap, not evidence of an existing user-input injection. An inline `` also passes, despite loading a new stylesheet permitted by the existing style policy. + +**Requested change:** parse HTML attributes with browser-compatible handling of unquoted values and character references, or explicitly reject unsupported loading syntax. Cover template-literal URLs in inline scripts and CSS loads in inline styles. Keep the URL list exact; do not widen it to a domain or wildcard. Add negative tests for these cases, with the existing positive fixture still passing. Static scanning cannot prove arbitrary JavaScript never constructs a URL, so describe its supported scope accurately. + +To reproduce the smallest case, add this test inside the existing `describe` block in `node-tests/docker-static-postbuild-test.js`: + +```js +it('rejects an unquoted unlisted CDN script', () => { + const body = ''; + assertFails(run(indexHtml({ body })), 'EXTERNAL_LOADS'); +}); +``` + +From the target repository root with dependencies installed, run: + +```sh +node node_modules/mocha/bin/mocha node-tests/docker-static-postbuild-test.js +``` + +At the reviewed head, the added test fails because the child process returns 0 instead of 1. After the fix, it must pass. Reviewer probes were kept outside the target source tree; the clone finished clean. + +### Checks that passed + +The two entries in `EXTERNAL_LOADS` are exact complete strings, including paths and query strings, compared with `includes`; there is no wildcard or domain-wide match. They are the existing MathJax 2.7.1 script and Open Sans stylesheet. The owner-approved SPEC-04 counted-only assets remain accepted scope; metrics adapter URL strings are not treated as proof of actual network loads. + +The production tag remains a synchronous classic script at the same insertion point. It now uses the build-time `assetsPrefix`, whose default is `/ember_osf_web/`. Production keeps `bundleZxcvbn: false` and excludes `zxcvbn.js` from filename fingerprinting. Addon 2.0.0 copies `require.resolve('zxcvbn/dist/zxcvbn.js')` into the public tree; the lockfile fixes zxcvbn at 4.4.2. Downloaded package archive SHA-1 values match the existing lockfile, and the extracted zxcvbn file matches the installed package. The old CDN file matches the removed SHA-256 SRI value. + +```toon +integrity[3]{artifact,check,value}: + zxcvbn-4.4.2.tgz,sha1,28ec17cf09743edcab056ddd8b1b06262cc73c30 + ember-cli-password-strength-2.0.0.tgz,sha1,f6f16d5c42ac16a4605305244f15f092f15224b3 + npm-zxcvbn-dist-file,sha256,f42c651f40506acb6b662490f338dd47a5951d3312039c4ab8fe5090484f351a +``` + +The script location and file come from trusted build configuration and dependencies, not password input, URL parameters, or the per-deployment renderer. The runtime renderer substitutes only backend configuration in the meta tag. nginx serves the static build file through both `/ember_osf_web/zxcvbn.js` and `/zxcvbn.js`; the final image runs as uid 101 and does not expose an upload handler for that root. No new user control over script bytes or source was found. + +`render.sh`, `default.conf`, and the lockfile are unchanged from the base. CSP still permits `'self'`, keeps the existing cdnjs/font sources, and adds no `unsafe-inline` or `unsafe-eval` to script policy. Existing renderer tests pass. Password service loading, asynchronous strength calculation, validator thresholds, and feedback handling are unchanged. Comparing the old CDN script and npm 4.4.2 in isolated JavaScript contexts produced identical results for ten inputs, including empty, weak, strength levels 0–4, Japanese text, and supplied user inputs, after excluding the measured `calc_time` field. + +### Validation and limits + +```toon +checks[5]{check,result,evidence}: + new_postbuild_tests,8_pass,local_Node_26.12.0_Mocha_6.2.1_Chai_4.2.0 + new_tests_with_base_postbuild,8_fail,7_exit_status_failures_and_1_log_text_failure + renderer_tests,166_pass,local_existing_suite + cdn_vs_npm_results,10_equal,all_result_fields_except_calc_time + existing_CI,success,run_43121865127_401_node_tests_pass_986_browser_tests_complete +``` + +The mutation run replaced only `postbuild.js` with the base revision, ran the new tests, and restored the file. Seven failures demonstrate missing guards; the positive test fails only because the old success message lacks `zxcvbn local`. These fixture tests do not independently prove that `ember-cli-build.js` emits the intended production tag. + +No production rebuild, live browser session, or Docker operation was run for this review. The reported 819639-byte production asset and before/after byte identity are author/task evidence, not an independently regenerated artifact. Source provenance, package integrity, behavior comparison, and existing CI were independently checked. The guard's `/zxcvbn/` file-content test is only a presence check, not a cryptographic integrity check. R1 and its regression tests remain required before a pass verdict. + +--- + + +## 日本語 + +[→ To English](#english) + +### 判定と範囲 + +**修正要求(changes requested)— P2 指摘 R1 が 1 件。** 変更された `ember-cli-build.js`、`docker/static/postbuild.js`、`node-tests/docker-static-postbuild-test.js` の全差分を確認した。固定依存、production ビルド段階、nginx、レンダラー、パスワード強度 addon と呼び出し元、既存 CI ログも確認した。報告前に head を再確認した。マージ・承認・コミット・push・スタック操作・runtime/.env の参照は行っていない。 + +### R1 — ブラウザが解釈する外部読み込みを検出してから完全一致で照合する + +**対象:** `docker/static/postbuild.js:186–205`、特に 199 行の `attributes(rawAttrs)` 呼び出し。98–109 行の補助関数は引用符付き属性しか読まない。新しい検査は引用符なし属性を拒否せず、HTML 文字参照も復号せずに流用している。186 行の URL 正規表現は直前のバッククォートも認識しない。補助関数自体は既存でも、新しい検査の保証に穴が生じている。 + +成功する fixture のローカル zxcvbn と既知 URL 2 件を残し、次を一つずつ追加した。 + +```html + + + +``` + +すべて終了コード **0**、`external loads: 2 listed, zxcvbn local` となる。同じ URL を通常の引用符付き `src` にすると正しく **1** となる。既存の `script-src https://cdnjs.cloudflare.com` がこれらの未掲載スクリプトを許し、インライン loader には CSP ハッシュまで生成される。これはビルド方針の回帰検出漏れであり、既存の利用者入力による注入を発見したという意味ではない。インラインの `` も通過し、既存の style 方針が許す新しい外部 CSS を追加できる。 + +**修正要求:** 引用符なし属性と文字参照をブラウザと整合する形で解析するか、未対応の読み込み構文を明示的に拒否する。インライン JavaScript のテンプレート文字列とインライン CSS の読み込みも扱う。URL 一覧は完全一致のままとし、ドメイン単位やワイルドカードへ広げない。各再現例の拒否テストを追加し、既存の正常 fixture は引き続き成功させる。任意の JavaScript が動的に URL を作らないことまでは静的走査で証明できないため、対応範囲も正確に記述する。 + +最小再現は `node-tests/docker-static-postbuild-test.js` の既存 `describe` 内に次を追加する。 + +```js +it('rejects an unquoted unlisted CDN script', () => { + const body = ''; + assertFails(run(indexHtml({ body })), 'EXTERNAL_LOADS'); +}); +``` + +依存を導入した対象リポジトリのルートから実行する。 + +```sh +node node_modules/mocha/bin/mocha node-tests/docker-static-postbuild-test.js +``` + +レビュー対象 head では、子プロセスが期待値 1 ではなく 0 を返すため追加テストが失敗する。修正後は成功する必要がある。レビュワの検証コードは対象ソースツリー外に保存し、クローンは差分なしで終えた。 + +### 合格した確認 + +`EXTERNAL_LOADS` の 2 件はパス・クエリを含む URL 全体を `includes` で比較しており、ワイルドカードやドメイン全体の許可はない。既存の MathJax 2.7.1 と Open Sans stylesheet に限る。所有者承認済みの SPEC-04 件数記録のみの外部資産は今回の許容範囲とし、metrics adapter 内の URL 文字列だけで実際の外部通信と判断していない。 + +production のタグは挿入位置を変えず、同期 classic script のままである。ビルド時の `assetsPrefix` を使い、既定値は `/ember_osf_web/`。production は `bundleZxcvbn: false` を維持し、`zxcvbn.js` はファイル名への指紋付与から除外される。addon 2.0.0 は `require.resolve('zxcvbn/dist/zxcvbn.js')` を public tree にコピーし、lockfile は zxcvbn を 4.4.2 に固定する。取得した配布アーカイブの SHA-1 は既存 lockfile と一致し、展開した zxcvbn は導入済みパッケージとも一致した。旧 CDN ファイルは削除された SHA-256 SRI 値と一致した。 + +```toon +integrity[3]{artifact,check,value}: + zxcvbn-4.4.2.tgz,sha1,28ec17cf09743edcab056ddd8b1b06262cc73c30 + ember-cli-password-strength-2.0.0.tgz,sha1,f6f16d5c42ac16a4605305244f15f092f15224b3 + npm-zxcvbn-dist-file,sha256,f42c651f40506acb6b662490f338dd47a5951d3312039c4ab8fe5090484f351a +``` + +読み込み先とファイルは信頼するビルド設定・依存由来であり、パスワード入力・URL パラメータ・起動時レンダラーから決まらない。起動時の置換対象は meta タグのバックエンド設定に限る。nginx は `/ember_osf_web/zxcvbn.js` と `/zxcvbn.js` の両方を同じ静的ファイルに解決する。最終イメージは uid 101 で動き、配信ルートへの upload handler はない。スクリプトの内容や読み込み元に新しい利用者操作経路は認めなかった。 + +`render.sh`、`default.conf`、lockfile は base と同一。CSP は `'self'` と既存の cdnjs・フォント元を維持し、script 方針に `unsafe-inline` や `unsafe-eval` を追加していない。既存レンダラー試験は成功した。パスワード service のロード、非同期の強度計算、validator の閾値、feedback の処理も無変更。旧 CDN と npm 4.4.2 を独立した JavaScript 実行環境で比較し、空文字・弱い文字列・強度 0–4・日本語・利用者情報を含む 10 入力で、計測値 `calc_time` 以外の全結果が一致した。 + +### 検証と限界 + +```toon +checks[5]{check,result,evidence}: + new_postbuild_tests,8_pass,local_Node_26.12.0_Mocha_6.2.1_Chai_4.2.0 + new_tests_with_base_postbuild,8_fail,7_exit_status_failures_and_1_log_text_failure + renderer_tests,166_pass,local_existing_suite + cdn_vs_npm_results,10_equal,all_result_fields_except_calc_time + existing_CI,success,run_43121865127_401_node_tests_pass_986_browser_tests_complete +``` + +差し戻し試験では `postbuild.js` だけを base 版へ一時置換し、新テストを実行してから復元した。7 件は検査がないため終了コードで失敗する。正常系 1 件は旧ログに `zxcvbn local` がないため失敗する。これらの fixture 試験だけでは、`ember-cli-build.js` が意図した production タグを生成することまでは独立に証明しない。 + +今回は production の再ビルド、実ブラウザ起動、Docker 操作を行っていない。819639 バイトの production 資産と変更前後のバイト一致は、作成者・依頼文の証拠であり、レビュワが再生成した成果物ではない。ソース由来、パッケージ整合性、動作比較、既存 CI は独立に確認した。ファイル内容の `/zxcvbn/` 検査は存在確認にすぎず、暗号学的な整合性検査ではない。合格判定には R1 の修正と回帰テストが必要である。 diff --git a/docs/reviews/2026-11-12_RDM-ember-osf-web-dev2027_PR4_Zxcvbn_Local_r2.md b/docs/reviews/2026-11-12_RDM-ember-osf-web-dev2027_PR4_Zxcvbn_Local_r2.md new file mode 100644 index 00000000..86f10872 --- /dev/null +++ b/docs/reviews/2026-11-12_RDM-ember-osf-web-dev2027_PR4_Zxcvbn_Local_r2.md @@ -0,0 +1,244 @@ +# REV PR4 Local zxcvbn review r2 / zxcvbn ローカル配信レビュー r2 + +> **Summary:** Changes requested: all r1 reproductions are fixed, but the replacement scanner still misses SVG, CSS and refresh loads and rejects some local resources as external. +> **概要:** 修正要求。r1 の再現例はすべて修正されたが、置換後の検査にも SVG・CSS・refresh の読み落としと、ローカル資源を外部読み込みとする誤検出がある。 + +**Jump:** [English](#english) · [日本語](#japanese) + +- Version / 版: 0.1 — 2026-11-12. **Learned:** parsing a document is insufficient if later processing discards namespaces or rewrites quoted strings as comments. / 文書を構文解析しても、その後で名前空間を捨てたり引用符内をコメントとして消したりすると読み込みの意味を失う。 +- Applies to: RCOSDP/RDM-ember-osf-web-dev2027 PR #4, head `310dfb883b167959e81e5b134bb315e971541054`; incremental review from `3e58dbc580b6c8e7ce617735e86c3bda11a7b6a9`. +- Reviewer / レビュワ: REV(担当レビュワ, Codex Astra); reviewed / レビュー日: 2026-11-12. +- Related: [Review index / レビュー索引](README.md) · [SPEC-04](../spec/04_acceptance_tests.md) · [SPEC-06](../spec/06_documentation_conventions.md). +- Used by / 利用先: PR #4 correction and re-review / PR #4 の修正・再レビュー、[review index / レビュー索引](README.md). + +--- + + +## English + +[→ 日本語版へ](#japanese) + +### Verdict and evidence + +**Changes requested — four P2 findings, R2–R5.** P2 means a defect to fix before accepting this build-policy guard. R1 is resolved for its four concrete reproductions: unquoted attributes, HTML character references, template literals and inline CSS imports now exit 1. This review covers the new `external-loads.js`, its integration in `postbuild.js`, the added tests and direct development dependencies. It does not reopen the accepted r1 provenance and runtime review. + +All **47** current postbuild tests pass. Replacing only the tested postbuild entry point with the r1 version produces **26 passes and 21 failures**, independently confirming the author's mutation result. There are 39 additions to the previous eight tests, rather than the 38 stated in the dispatch. Tests used Node 26.12.0 and isolated dependencies matching the lockfile's new direct entries: parse5 4.0.0, postcss 6.0.23 and @babel/parser 7.6.2; Mocha 6.2.1 and Chai 4.2.0 supplied the test runner. This was not a full frozen-lockfile install or a production rebuild. + +The examples in R2–R5 were added individually to the existing passing `indexHtml()` fixture, preserving both allowed URLs, the local zxcvbn tag and its file. Each was run through the real `postbuild.js` child process, not only the extraction helper. R2–R4 exit **0** with `external loads: 2 listed, zxcvbn local`; they should be rejected. R5 exits **1** for resources that do not introduce an external network fetch. Browser behavior is inferred from the cited standards and the inspected CSP (Content Security Policy, the browser's rules for permitted resources); no live browser network capture was performed. + +### R2 / P2 — Preserve SVG attribute namespaces + +**Location:** `docker/static/external-loads.js:245–249`. + +```html + +``` + +parse5 returns two attributes named `href`; the first also has `prefix: "xlink"` and a namespace. The new attribute map discards that information and keeps the first value, so it sees only `/local.svg`. SVG instead gives the unnamespaced `href` precedence regardless of attribute order. Reversing the two attributes makes the guard reject the same external load. The equivalent `` is also missed. SVG's precedence rule is explicit in [SVG 2 linking](https://www.w3.org/TR/SVG2/linking.html#XLinkRefAttrs). + +The `image` example is sufficient for impact: the current `img-src` allows HTTPS images, so CSP does not close this hole. Keep namespace-qualified attributes distinct and apply browser precedence, or inspect both conservatively. Add both attribute orders to regression tests. + +### R3 / P2 — Do not remove comment markers inside CSS strings + +**Location:** `docker/static/external-loads.js:102–122`, especially line 112. + +```html + +``` + +postcss parses the declaration, but `cssTextUrls()` then applies `CSS_COMMENT` to its raw text. It removes everything between the `/*` and `*/` inside two quoted URL strings, including the middle external candidate. The same declaration in a `style` attribute also passes. These are URL strings, not CSS comments: [CSS Images 4](https://www.w3.org/TR/css-images-4/#image-set-notation) defines image-set strings as URLs and permits the browser to select a candidate. Existing `style-src 'unsafe-inline'` and `img-src https:` permit this resource path. + +Use CSS value tokens, or a scanner that preserves string and escape state, for comment handling and URL extraction. Test this case in both ` +``` + +postcss は宣言を解析するが、続く `cssTextUrls()` は生のテキストへ `CSS_COMMENT` を適用する。引用符付き URL 二つの中にある `/*` と `*/` の間を消すため、中央の外部候補も消える。同じ宣言を `style` 属性に置いても通過する。これらは CSS コメントではなく URL 文字列である。[CSS Images 4](https://www.w3.org/TR/css-images-4/#image-set-notation) は image-set の文字列を URL と定め、ブラウザが候補を選択できる。現行 `style-src 'unsafe-inline'` と `img-src https:` はこの読み込みを許す。 + +CSS の値をトークンに分けて扱うか、文字列・エスケープ状態を保持する走査器でコメントと URL を処理すること。`` | `https://assets.example.test/a.css` | +| `` | `https://assets.example.test/a.png` | +| `` | first/external src is used | +| `` | with empty root context: `http://assets.example.test/a.js` | + +The contextPath fixture was rendered by substituting the empty root-context value, not by running a JSP engine; `etc/jetty/dist-context.xml` configures `/`, so empty servlet contextPath is relevant. The scanner explicitly trusts this expression but never rechecks the suffix when substituted. These cases are ordinary static markup/CSS or a specifically permitted JSP expression, not the accepted limit of arbitrary runtime-assembled JavaScript. + +Decode attribute references and normalize URLs using browser-compatible rules; reject ambiguous/duplicate resource attributes; tokenize CSS including comments and escapes; and validate the entire context-relative expression after substitution, including slash/backslash suffixes. Add the listed fixtures. Literal JavaScript `fetch("https://assets/data")` in a `.js` file also exits 0 because JS_URL requires a dotted host; a single-label host is still another host and should not escape a literal-URL rule. + +### Checks that now work and accepted limits + +Independent repetitions reject the r2 uppercase, spaces-around-equals, multiline tag, srcset, uppercase inline CSS URL, unknown EL variable/composed scheme, and space-containing filename fixtures. Protocol-relative `//host`, `\\host` and `/\host` attributes are rejected. Actual `.css` @import and `.js` literal fetch fixtures are rejected; the two raw text versions placed outside style/script in a JSP are not browser loads and are not regression findings. The supplied 17 negative and one clean fixtures run before the real WAR and pass. A corrupt archive returns 2; a stale reviewed-list entry returns 1. Zip entry iteration removes the shell word-splitting/error-suppression problem. + +`url(/**/https://assets.example.test/a.png)` was specifically tested: Chrome treated the comment-looking text inside this unquoted URL as part of a same-origin path, not as an external request. It is **not** reported as an external-load bypass. CSS import comments and escapes in the table are actual external loads. + +A literal `data-src="https://assets.example.test/a.png"` copied into `img.src` by JavaScript also passes and triggers an external browser request. This is indirect runtime data flow rather than URL assembly from fragments. The Java comment and PR body document only fragment assembly; extend that limitation to attribute/DOM/configuration-derived destinations and ensure rendered-page coverage exercises them. General runtime-assembled URLs are acceptable as a documented static-analysis limit under the review instructions; the static examples in R3-01 still need fixes. No exhaustive JavaScript data-flow proof is requested. + +### Logo safety and compatibility observation + +Both login and OTP now drop ordinary absolute logo URLs and show the service name as fallback; no new executable HTML sink is added. However, the actual CAS 4.1.5 `AbstractRegisteredService.getLogo()` returns `java.net.URL` (verified from the built WAR's class), not a root-relative string. Its string form contains a scheme and will not start with `/images/` or `/themes/`. Existing service JSON files use `http://localhost:8080/images/...`, which this predicate hides even when it identifies the CAS's own image. The change is fail-closed, but effectively disables service logos rather than retaining the stated local-logo support. This is a compatibility observation, not an attacker-controlled-logo exploit claim. + +Either explicitly document intentional removal of all service logos, or provide a validated local path model/normalization and tests for a real RegisteredService. If local string rendering is added, escape the final attribute value: c:url rewrites URLs, it is not an HTML attribute encoder. A synthetic injected-srcset example was explored, but because the actual property is URL-typed and rejected by the current prefix test, it is not presented as a reachable exploit in this head. + +### Validation and evidence + +The checkout was streamed into an isolated container using the exact Dockerfile-pinned Maven/Temurin 8 digest. `mvn -B package -DskipTests=false` succeeded in 1:31, OAuth 109 and OSF 45 tests, with zero failures/errors/skips. `sh etc/check-war-assets.sh cas-server-webapp/target/cas.war` reports 18 expected fixture results and 57 clean served files. Other checks disabled by the default nocheck profile are not claimed. The built WAR is retained as `../rev-cas5-r3.war`; all reviewer containers were removed; git diff check passes and the source checkout is clean. + +Evidence: [build](../rev-cas5-r3-build.log), [fixture inputs](../rev-cas5-r3-mutants/cases.json), [scanner outcomes](../rev-cas5-r3-mutants.log), [browser harness](../rev-cas5-r3-browser.cjs), [browser outcomes](../rev-cas5-r3-browser.log), [extra cases and getLogo signature](../rev-cas5-r3-extra.log). Browser fixtures use installed headless Chrome, fulfill only their synthetic same-origin document, and abort every resource request. The author's S5 2-pass and fakecas 22-pass reruns were supplied in the task; they were not independently rerun or audited as new run folders here. No real server-rendered JSP, end-to-end login or deployed CSP was exercised in r3. + +--- + + +## 日本語 + +[→ To English](#english) + +### 判定 + +**要修正 — R3-01 / P2。R2-01 は改善したが未完了。** 既存アプリの外部資産除去は維持されている。指摘は新ビルド検査の保証についてであり、現在の通常ログイン画面に外部ライブラリが残るという主張ではない。ソース変更・merge・approve・push・commit・grdm-dev スタック操作は行っていない。 + +### R3-01: 検査を通る静的構文がブラウザでは外部を読む + +`etc/war-asset-check/CheckWarAssets.java:241–259` は HTML 文字参照を復号せず、URL が無視する制御文字も正規化せずに文字列を検査する。`css():222–229` は CSS のコメント・トークン・escape を扱わない。`attributes()` は同名属性の最後の値で map を上書きするが、HTML ブラウザは最初の重複属性を使う。英語節の7例は単独 WAR ですべて終了0となり、headless Chrome が右欄の外部要求を試みた。要求は全件捕捉してネットワークに出る前に中止した。 + +具体的には `https:`、`//`、`ht tps`、`@import/**/`、CSS の `https\3a `、外部 src の後にローカル src を重複指定する形、`${pageContext.request.contextPath}//host` が通る。contextPath の例は JSP エンジンではなく root context の空文字を代入した fixture で確認した。配備設定 `etc/jetty/dist-context.xml` は `/` なので空の servlet contextPath は対象環境にも関係する。検査器は式を許可するだけで、置換後の後続文字列を再検査していない。これらは通常の静的 HTML/CSS または明示的に許可した JSP 式であり、許容された任意 JavaScript の実行時組み立てという限界ではない。 + +属性の文字参照の復号とブラウザに沿った URL 正規化、重複資産属性の拒否、コメント・escape を含む CSS のトークン解析、contextPath の置換後の全体検証(slash/backslash の後続も含む)、負例追加が必要。`.js` 内の `fetch("https://assets/data")` も JS_URL がドット付きホストを必須にするため終了0となる。単一ラベルのホストも別ホストなので、直書き URL の検査対象から漏らしてはならない。 + +### 改善を確認した範囲と許容する限界 + +前回の大文字、属性前後の空白、複数行タグ、srcset、大文字 inline CSS URL、未知の EL 変数・scheme 連結、空白入りファイル名を独立再試験して拒否を確認した。`//host`・`\\host`・`/\host` の属性も拒否する。実 `.css` の @import、実 `.js` の直書き fetch も拒否した。JSP の style/script 外に単なる文字列として置いた2例はブラウザ読み込みではなく、回帰として数えていない。同梱17負例・1正常例は実 WAR より前に実行され合格。壊れた archive は終了2、古くなったレビュー一覧項目は終了1。zip を直接読むため前回のシェル単語分割・エラー隠蔽は解消した。 + +`url(/**/https://assets.example.test/a.png)` は Chrome ではコメント風文字列を同一 origin のパスの一部と解釈し、外部要求にはならなかった。したがって外部読み込みの回避例には含めない。英語表の import コメントと CSS escape は実際に外部要求になる。 + +`data-src="https://assets.example.test/a.png"` を JavaScript が `img.src` にコピーする例も検査を通り、外部要求になる。これは断片の連結ではなく実行時の間接的データ移動。Java コメントと PR 本文は断片組み立てだけを限界としているため、DOM 属性・設定由来の宛先も含めて限界を記載し、生成画面試験で該当経路を通すこと。指示どおり一般の実行時組み立ては文書化された限界として許容するが、R3-01 の静的な例は修正が必要。JavaScript の全データフローの証明を求めるものではない。 + +### ロゴの安全性と互換性上の観察 + +ログイン・OTP とも通常の絶対 URL のロゴを表示せず、サービス名へフォールバックする。新しい実行可能 HTML sink は追加していない。ただし完成 WAR の CAS 4.1.5 クラスで確認すると `AbstractRegisteredService.getLogo()` は root-relative な String ではなく `java.net.URL` を返す。文字列表現は scheme を含み、`/images/`・`/themes/` で始まらない。既存 JSON の `http://localhost:8080/images/...` も、CAS 自身の画像を示す場合さえ非表示になる。安全側に閉じる変更だが、説明されたローカルロゴ維持ではなく、事実上サービスロゴをすべて無効にする。これは互換性の観察であり、攻撃者がロゴを操作できるという主張ではない。 + +全ロゴの廃止を意図するなら明記するか、ローカルパスとして検証した値を持つモデル・正規化と実 RegisteredService を使う試験を用意すること。今後ローカル文字列を出す場合は最終属性値を escape すること。c:url は URL の書き換えであり HTML 属性エンコーダではない。srcset 注入の模擬例も調査したが、現実の URL 型と現行の先頭判定によりその文字列は受理されないので、この head で到達可能な exploit としては扱わない。 + +### 検証と証拠 + +Dockerfile 固定の Maven/Temurin 8 digest の隔離コンテナに checkout を転送し、`mvn -B package -DskipTests=false` が1分31秒で成功。OAuth 109件・OSF 45件、失敗・エラー・skip なし。`sh etc/check-war-assets.sh cas-server-webapp/target/cas.war` は18 fixture が期待どおり、57配布ファイルは所見なしと報告した。既定 nocheck profile が無効にする他の検査の合格は主張しない。WAR は `../rev-cas5-r3.war` に保存し、レビュワのコンテナは全削除。git diff check 合格、ソース checkout は clean。 + +証拠: [build](../rev-cas5-r3-build.log)、[入力](../rev-cas5-r3-mutants/cases.json)、[検査結果](../rev-cas5-r3-mutants.log)、[browser harness](../rev-cas5-r3-browser.cjs)、[browser 結果](../rev-cas5-r3-browser.log)、[追加試験と getLogo 型](../rev-cas5-r3-extra.log)。headless Chrome は模擬同一 origin 文書だけを返し、資産要求をすべて中止した。作者の S5 2合格・fakecas 22合格はタスクで提示された結果であり、今回独立再実行や新規実行フォルダーの監査はしていない。r3 では実 JSP 描画・ログイン全体・稼働 CSP を試験していない。 diff --git a/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r4.md b/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r4.md new file mode 100644 index 00000000..e68d5e9d --- /dev/null +++ b/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r4.md @@ -0,0 +1,86 @@ +# REV CAS PR5 external asset regression guard r4 / CAS PR5 外部資産回帰検査レビュー r4 + +> **Summary:** PASS within the explicitly narrowed regression-guard scope. The r3 concrete cases and ordinary external resource additions are rejected; deliberate obfuscation and runtime data flow remain documented limits. +> **概要:** 明示された回帰検査の範囲で合格。r3 の具体例と通常の外部資産追加を拒否し、意図的難読化と実行時データフローは限界として文書化されている。 + +**Jump:** [English](#english) · [日本語](#japanese) + +- Version / 版: 0.1 — 2026-11-16. **Learned:** a bounded build regression check is useful when concrete failures are tested and runtime observations are kept distinct from static guarantees. / 具体的な失敗例を試験し、実行時の観察と静的保証を区別すれば、範囲を限定したビルド回帰検査は有用である。 +- Applies to: RDM-cas-overlay-dev2027 PR #5 `34a5ed6242f2233f3ea01ee70d79bb2749e6c418`, delta from `878099b0405060a0d05c0d52cb286e53abf381e1`. +- Posted comment / 投稿コメント: issuecomment-6902722689. +- Reviewer: REV(担当レビュワ, Codex Astra). +- Related: [Review index](../rev-guide88/docs/reviews/README.md) · [SPEC-06](../rev-guide88/docs/spec/06_documentation_conventions.md) · [r3](2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r3.md). Used by: CAS integration and guide PR #88 pin update. + +--- + + +## English + +[→ 日本語版へ](#japanese) + +### Verdict and scope + +**PASS; no blocking finding within the coordinator's revised scope. R3-01's concrete cases are resolved.** The scanner is a build-time regression guard for accidentally added external assets, not a security proof against deliberate obfuscation. Runtime page requests are checked separately by the guide's loopback guard. No GitHub approval, merge, push, commit or grdm-dev stack operation was performed. + +### Independent regression tests + +I recompiled the Java scanner in the Dockerfile's exact Maven/Temurin 8 image and ran 28 independent WAR fixtures: 27 negative cases returned 1, and the local-script control returned 0. These include every blocking r3 example: numeric entity scheme/slashes, tab in scheme, CSS import with a comment, CSS hexadecimal escape, duplicate src with external first/local second, contextPath followed by a protocol-relative host, and a single-label host in a JS literal. Existing uppercase, multiline, spaces-around-equals, unknown EL, srcset, mixed slash/backslash and filename-with-space cases remain rejected. + +Ordinary, non-obfuscated additions also fail as intended: external script src; link rel=stylesheet href; img src; CSS background url(); quoted and url()-form @import; @font-face src; a style block; inline style; and picture/source srcset with multiple candidates. No plain accidental external asset false negative was found in this matrix. The supplied fixtures additionally cover named entities, percent encoding and contextPath/backslash. The artifact guard checks duplicate occurrences instead of overwriting them, validates the contextPath suffix, and retains failure on scanner errors. + +Evidence: [independent inputs](../rev-cas5-r4-mutants/cases.json) and [scanner results](../rev-cas5-r4-mutants.log). This is finite regression evidence, not a claim to recognize all possible HTML/CSS/JavaScript semantics. Further adversarial encodings outside the stated normalizations were deliberately not pursued under the revised assignment. + +### Documentation and compatibility + +The Java header and latest PR-body correction explicitly state the accidental-regression purpose, disclose runtime assembly and attribute/DOM/configuration-derived URLs (including data-src), and acknowledge unknown named references and CSS tokenization corner cases. The PR explicitly supersedes the stronger historical wording. Guide SPEC-04 0.19, inspected at PR #88 head `12d633e4170833e570dd24220f481d35d11c800e`, makes the same distinction in English and Japanese. This is an honest bounded description for the reviewed scope. + +Runtime qualification: the guide guard aborts off-loopback requests, but existing KNOWN_EXTERNAL_ASSETS are counted rather than failed. The guide documents that exception. Thus “zero CAS external requests” is a live observation to maintain, not a theorem inferred from any green aggregate run or a claim that the runtime guard automatically fails every known-asset request from CAS. The live counts are author-supplied evidence; no new live run was performed here, and runtime assurance covers exercised paths only. + +Service logos remain entirely hidden because getLogo() returns an absolute java.net.URL. The PR body and reviewed-dynamic-urls.tsv now explicitly acknowledge this, including CAS-local absolute logos. That accepted, nonblocking compatibility change does not get reclassified as a new blocker. No view or application-source change was made in this round; only the scanner, its fixtures and its reviewed-list explanation changed. + +### Build and artifact verification + +`mvn -B package -DskipTests=false` succeeded in 1:35 in the Dockerfile-pinned image: OAuth 109 and OSF 45 tests (154 total), no failures/errors/skips. The shipped shell entrypoint passed all 30 fixtures (29 bad, one clean) and scanned 57 served files without findings. The default nocheck profile still disables other static checks, which are not claimed. + +SHA-256 comparison against my retained r3 WAR found all 57 served-source files identical. Among all 557 non-library file entries, only `META-INF/maven/io.cos.cas/cas-server-webapp/pom.properties` and `WEB-INF/classes/META-INF/MANIFEST.MF` differ in build timestamps; no files were added or removed. This uses a different enumeration from the author's 207-file comparison and does not reproduce its exact count, but independently confirms unchanged application/served content. + +No new browser/JSP or live S5/full-suite run was performed. Reusing the previous live results is reasonable for unchanged served assets, while retaining their prior coverage limits. The independently built artifact comparison does not certify arbitrary runtime configuration, every possible route, or overall deterministic suite behavior. Guide #88 still requires its separate dependency-pin integration; this review does not perform it. + +Logs: [Maven and shipped guard](../rev-cas5-r4-build.log), [artifact comparison](../rev-cas5-r4-artifact-compare.json). The WAR is retained at `../rev-cas5-r4.war`. The source checkout remains unmodified; `git diff --check 878099b0..HEAD` passes, and reviewer containers were removed. + +--- + + +## 日本語 + +[→ To English](#english) + +### 判定と範囲 + +**合格。調整された範囲で阻害指摘なし。R3-01 の具体例は解消。** 検査器はうっかり足された外部資産に対するビルド時の回帰検査であり、意図的難読化に対する安全性の証明ではない。実行時の画面要求はガイドの loopback guard で別途検査する。GitHub の承認・merge・push・commit・grdm-dev スタック操作は行っていない。 + +### 独立した回帰試験 + +Dockerfile 固定の Maven/Temurin 8 イメージで Java 検査器を再コンパイルし、独立 WAR fixture 28件を実行した。負例27件は終了1、ローカル script の対照1件は終了0。r3 の阻害例である数字文字参照の scheme/slash、scheme 中の tab、コメント付き CSS import、CSS 16進 escape、外部→ローカルの重複 src、contextPath に続くスキーム省略ホスト、JS の単一ラベルホストをすべて含む。従来の大文字・複数行・等号前後の空白・未知 EL・srcset・slash/backslash 混在・空白ファイル名も引き続き拒否した。 + +難読化していない通常の追加でも、script src、link rel=stylesheet href、img src、CSS background url()、引用符形式と url() 形式の @import、@font-face src、style 要素、inline style、picture/source の複数候補 srcset を正しく拒否した。この試験集合では通常の誤追加を見逃す例はなかった。同梱 fixture は名前付き文字参照・パーセント符号・contextPath/backslash も検査する。重複属性を上書きせず全件調べ、contextPath の後続を検証し、検査エラー時も失敗させる。 + +証拠は [独立入力](../rev-cas5-r4-mutants/cases.json) と [結果](../rev-cas5-r4-mutants.log)。有限個の回帰試験であり、HTML/CSS/JavaScript の全構文を理解するという主張ではない。今回の指示に従い、明示された正規化の外にある追加の攻撃的符号化は追求していない。 + +### 文書と互換性 + +Java ヘッダと最新の PR 本文訂正は誤追加の回帰防止という目的を明記し、実行時組み立て・属性/DOM/設定由来の URL(data-src を含む)、未知の名前付き文字参照、CSS 字句解析の端の事例を限界として記載する。PR は過去の強い表現を明示的に訂正した。PR #88 head `12d633e4170833e570dd24220f481d35d11c800e` の SPEC-04 0.19 も英和両方で同じ区別をする。今回の範囲に対して限界を隠さない説明になっている。 + +実行時についての限定: ガイドの guard は loopback 外の要求を中止するが、既存 KNOWN_EXTERNAL_ASSETS は失敗ではなく件数を記録する。この例外はガイドにも記載される。したがって「CAS の外部要求0」は維持すべき実機観察であり、集計結果が緑なら自動的に成り立つ証明でも、既知資産要求が CAS から出れば必ず試験が失敗するという主張でもない。実機件数は作者提供の証拠で、今回再実行していない。実行時保証は通った経路の範囲に限られる。 + +getLogo() が絶対 java.net.URL を返すため、サービスロゴはすべて非表示のまま。PR 本文と reviewed-dynamic-urls.tsv は CAS 自身を示す絶対 URL も含め、この互換性影響を明記した。受け入れられた非阻害事項を新しい阻害事項にはしない。今回変更したのは検査器・fixture・レビュー一覧の説明だけで、画面やアプリのソースは変更していない。 + +### ビルドと成果物検証 + +Dockerfile 固定イメージで `mvn -B package -DskipTests=false` が1分35秒で成功した。OAuth 109件・OSF 45件、合計154件が失敗・エラー・skip なし。出荷する shell entrypoint は30 fixture(負例29、正常1)に合格し、57配布ファイルを所見なしと判定した。既定 nocheck profile が無効にする他の静的検査の合格は主張しない。 + +保存したレビュワの r3 WAR と SHA-256 比較し、57配布ソースは全件同一だった。ライブラリ以外の全557ファイルでは `META-INF/maven/io.cos.cas/cas-server-webapp/pom.properties` と `WEB-INF/classes/META-INF/MANIFEST.MF` のビルド時刻だけが異なり、追加・削除はない。作者の207件とは列挙条件が異なるためその件数を再現したものではないが、アプリ・配布内容の不変を独立確認できた。 + +新たな browser/JSP 描画・実機 S5/full suite は実行していない。配布資産が変わらない範囲で前回実機結果を利用することは妥当だが、以前の検証範囲の限界は維持する。成果物比較は任意の実行時設定・全経路・スイート全体の決定性を証明しない。ガイド #88 の依存先固定版の統合は別途必要であり、このレビューでは行っていない。 + +ログは [Maven と出荷検査](../rev-cas5-r4-build.log)、[成果物比較](../rev-cas5-r4-artifact-compare.json)。WAR は `../rev-cas5-r4.war` に保存。ソース checkout は未変更、`git diff --check 878099b0..HEAD` は合格、レビュワのコンテナは削除した。 diff --git a/docs/reviews/2026-11-16_RDM-developer-guide_PR85_Health_Settle_r1.md b/docs/reviews/2026-11-16_RDM-developer-guide_PR85_Health_Settle_r1.md new file mode 100644 index 00000000..b8b579fb --- /dev/null +++ b/docs/reviews/2026-11-16_RDM-developer-guide_PR85_Health_Settle_r1.md @@ -0,0 +1,62 @@ +# PR85 Health settle review r1 / 健康判定の待機レビュー r1 + +> **Summary:** Request changes: unknown uptime bypasses the new readiness condition. +> **概要:** 修正要求。稼働時間が不明でも新しい健康判定の条件を通過する。 + +**Jump:** [English](#english) · [日本語](#japanese) + +- Version / 版: 0.1 — 2026-11-16. **Learned:** unknown uptime and proven long uptime need distinct outcomes. / 不明な稼働時間と、十分長いと確認できた稼働時間は区別する必要がある。 +- Applies to: RCOSDP/RDM-developer-guide PR #85, head `e5bf62263dc0595e7acd4179a8f61a706249af77`, base `6b7cadb625e44f93fd5d7ce5d251f4aceada284d`. +- Related: [PR #85](https://github.com/RCOSDP/RDM-developer-guide/pull/85) · [Review index](../rev-pr85/docs/reviews/README.md) · [SPEC-02](../rev-pr85/docs/spec/02_installer_cli.md) · [SPEC-06](../rev-pr85/docs/spec/06_documentation_conventions.md) + +--- + + +## English + +[→ 日本語版へ](#japanese) + +### Verdict and finding + +**Request changes — R1-01 (P2): fail closed for unknown uptime.** In `installer/grdm_dev/health.py:179–184`, `None` from `uptime_seconds` preserves `ok=True`. A row with `Service=worker`, `State=running`, `Health=""` and missing or empty `Status` passes `check_services` despite having no evidence of 20 seconds of continuous uptime. The added success test explicitly accepts empty Status. This contradicts SPEC-02 0.41 and leaves the original false-success path open when the field is absent or unrecognized. Parse recognized minute/hour/day forms separately from unknown data; reject missing, malformed, or non-Up Status and test those rejection cases. This is a reproduced contract gap; the observed real row did contain Status. + +### Parser and callers + +- `Up Less than a second`, `Up 1 second`, and `Up 19 seconds` return 0, 1, and 19 and fail; `Up 20 seconds` passes. Seconds with `(healthy)` or `(health: starting)` suffixes parse correctly. A real `Health=starting` still fails; `Health=healthy` intentionally bypasses settling. +- `Up About a minute`, `Up N minutes`, and `Up 3 hours` return None and pass. Their acceptance is appropriate at a fixed 20-second threshold, but the parser cannot distinguish them from absent or invalid Status. `Restarting (1) …`, `Exited …`, and `Created` also return None; matching non-running State correctly rejects them. An inconsistent running State would pass, as confirmed with synthetic rows, not observed in Docker. +- The sole real Docker command was the requested `docker compose -p grdm-dev ps --all --format json | head -c 2000`. The complete admin row had `Service=admin`, `State=running`, `Health=""`, and `Status="Up 2 hours"`; capitalized Compose JSON keys match the code. The captured row labels identify Compose 5.6.0, so it is not an independent historical v2 runtime test. Array/NDJSON parsing and lowercase aliases are supported in the source. +- `lifecycle.up` and secrets rotation both use `wait_until_healthy` (5-second polling; default timeouts 1800/5400 and 300 seconds respectively). They wait through settling; a deliberately short timeout can fail as expected. +- Acceptance preflight and scan check once. Immediately after a **successful** `up`, unchanged services have already met the same uptime threshold, so there is no new spurious failure. A restart after that success, or invoking them immediately after raw Compose startup, legitimately fails. This does not prove that a service surviving 20 seconds will never crash later. +- Tests cover rejection below the boundary, acceptance at the boundary and long uptime, explicit healthy bypass, and restarting State. Missing/unknown Status rejection is missing and the empty-string success assertion enshrines R1-01. SPEC-02 English and Japanese agree; version 0.40 and older entries remain. + +### Validation + +Read `gh pr diff 85` and used a fresh detached `rev-pr85` worktree. No source edits, commits, pushes, approvals, or merges were made. + +Installer suite: 724 tests passed, 13 skipped, 65.005 seconds on local Python 3.15.2. Ran unittest discovery with PYTHONPATH=installer and a subprocess guard that skips real Docker invocations, preserving the single-command restriction (the Compose config test is excluded). An initial attempt without PYTHONPATH failed imports (87 tests, 21 errors, 2 skips); the corrected full run supersedes it. `tools/docs_check.py` passed: 218 files, 8 exempt, 67 warnings. `tools/manifest_check.py` passed with zero findings. Logs: `/tmp/pr85-installer-guarded.log`, `/tmp/pr85-docs-check.log`. All 13 PR CI checks passed, including Python 3.11/3.12 unit tests, docs build/links, manifest policy, gitleaks, Semgrep, Trivy, shell checks, and platform dry runs. The unit jobs were initially pending and had finished at the final snapshot; no long polling was used. + +--- + + +## 日本語 + +[→ To English](#english) + +### 判定と指摘 + +**修正要求 — R1-01(P2):稼働時間が不明なら安全側に失敗させる。** `installer/grdm_dev/health.py:179–184` は `uptime_seconds` が None を返すと `ok=True` を保つ。`Service=worker`、`State=running`、`Health=""` で Status が欠落または空の行は、20 秒続けて動いた証拠がなくても `check_services` を通る。追加された成功テストも空の Status を明示的に許容している。SPEC-02 0.41 と一致せず、フィールドの欠落や未知の形式で元の誤成功の経路が残る。既知の分・時間・日の形式と不明な値を区別し、欠落・不正・Up 以外の Status を拒否するテストを追加してほしい。再現済みの契約上の穴であり、実機で得た行には Status が存在した。 + +### 解析と呼び出し元 + +- `Up Less than a second`、`Up 1 second`、`Up 19 seconds` は 0・1・19 となり拒否し、20 秒は通る。秒数に `(healthy)`・`(health: starting)` が付いても解析できる。実際の Health が starting なら拒否、healthy なら意図どおり待機を省く。 +- `Up About a minute`、`Up N minutes`、`Up 3 hours` は None となり通る。固定の20秒の閾値では妥当だが、不明・欠落と区別できない。`Restarting (1) …`、`Exited …`、`Created` も None となるが、対応する非 running の State で拒否される。State が running と矛盾すれば通ることは合成した行で確認したが、実機では観測していない。 +- 実機 Docker の実行は指定の `docker compose -p grdm-dev ps --all --format json | head -c 2000` 一回。完全に読めた admin 行は `Service=admin`、`State=running`、`Health=""`、`Status="Up 2 hours"` で、大文字始まりの JSON キーは実装と合う。ラベルは Compose 5.6.0 なので、過去の v2 を独立に実機試験したわけではない。ソースは JSON 配列・一行一オブジェクトと小文字の別名を扱う。 +- lifecycle の up と秘密の更新は共に `wait_until_healthy` を使う。5 秒ごとに調べ、既定の上限はそれぞれ1800/5400秒と300秒。待機を吸収し、利用者が特に短い上限を指定すれば失敗し得る。 +- accept の事前検査と scan は一回だけ調べる。**成功した** up の直後なら、変化していないサービスは同じ閾値を既に満たすため、新たな誤失敗は起きない。その後の再起動や Compose で直接起動した直後なら、拒否は妥当。20秒動いたサービスが後で異常終了しない保証ではない。 +- テストは境界未満の拒否、境界・長時間の成功、healthy の免除、restarting の拒否を扱う。欠落・未知の Status を拒む試験はなく、空文字の成功を期待する試験が R1-01 を固定している。SPEC-02 の英和は一致し、0.40 以下の履歴も残っている。 + +### 検証 + +`gh pr diff 85` を読み、新しい detached の `rev-pr85` worktree で確認した。ソース変更・コミット・push・承認・マージは行っていない。 + +ローカル Python 3.15.2 の installer 試験は724件成功、13件スキップ、65.005秒。PYTHONPATH=installer を設定して unittest discovery を実行し、実 Docker 呼び出しは subprocess のガードでスキップした(Compose config 試験を除外し、Docker 一回の制限を維持)。最初の PYTHONPATH なしの試行は import に失敗した(87件、21エラー、2件スキップ)が、修正後の全体試験は成功。tools/docs_check.py は218ファイル、8免除、67警告で成功。tools/manifest_check.py は指摘0件で成功。ログは `/tmp/pr85-installer-guarded.log` と `/tmp/pr85-docs-check.log`。PR CI は全13項目成功し、Python 3.11/3.12、文書ビルド・リンク、manifest 方針、gitleaks、Semgrep、Trivy、shell 検査、OS別 dry run を含む。初回は単体試験が実行中だったが最終確認で完了し、長時間のポーリングはしていない。 diff --git a/docs/reviews/2026-11-16_RDM-developer-guide_PR85_Health_Settle_r2.md b/docs/reviews/2026-11-16_RDM-developer-guide_PR85_Health_Settle_r2.md new file mode 100644 index 00000000..5c15e024 --- /dev/null +++ b/docs/reviews/2026-11-16_RDM-developer-guide_PR85_Health_Settle_r2.md @@ -0,0 +1,60 @@ +# PR85 Health settle review r2 / 健康判定の待機レビュー r2 + +> **Summary:** R1-01 is resolved; no blocking findings in the revised implementation. +> **概要:** R1-01 は解消した。修正後の実装に阻害事項は見つからなかった。 + +**Jump:** [English](#english) · [日本語](#japanese) + +- Version / 版: 0.1 — 2026-11-16. **Learned:** explicit unknown-uptime rejection preserves the same readiness contract across polling and one-shot callers. / 稼働時間不明を明示的に拒否すれば、繰り返す呼び出し元と一回だけの呼び出し元で同じ健康判定の契約を保てる。 +- Applies to: PR #85, head `5a0733c252a86a2dd2f139b8d75b7cd045c2bcf4`, base `6b7cadb625e44f93fd5d7ce5d251f4aceada284d`. +- Related: [PR #85](https://github.com/RCOSDP/RDM-developer-guide/pull/85) · [r1](2026-11-16_RDM-developer-guide_PR85_Health_Settle_r1.md) · [Review index](../rev-pr85/docs/reviews/README.md) · [SPEC-02](../rev-pr85/docs/spec/02_installer_cli.md) · [SPEC-06](../rev-pr85/docs/spec/06_documentation_conventions.md) + +--- + + +## English + +[→ 日本語版へ](#japanese) + +### Verdict + +**PASS, R1-01 resolved; no new blocking finding.** Missing, empty, and unrecognized Status now set `ok=False` and `health="uptime unknown"`. Recognized minutes, hours, days, and weeks are converted to seconds instead of sharing the unknown result. The added rejection tests include missing Status, empty Status, `Up for a while`, and `Created`; success cases include the 20-second boundary, approximate minute/hour, minutes, hours, days, and weeks. Existing short-uptime rejection, health suffix parsing, explicit healthy bypass, and restarting rejection remain covered. SPEC-02 0.41 English and Japanese both describe unknown uptime rejection; earlier version entries remain intact. + +### Caller verification + +Inspected accept's `default_health_check`, scan's `services_healthy`, and the shared `wait_until_healthy` used by lifecycle up and secrets rotation. Ran an independent injected-runner exercise through the actual accept and scan wrappers: missing/empty/unknown Status and 19 seconds reject; 20 seconds, About a minute, About an hour, 2 days, and 3 weeks pass. A separate 19→20-second sequence passes the polling helper. These checks used temporary Compose JSON and no Docker or external HTTP calls. + +One-shot accept/scan checks immediately after a successful up do not acquire a new waiting period: up has already satisfied the same threshold. A restart since up, direct Compose startup before 20 seconds, or missing uptime evidence correctly fails. Explicitly healthy and successful one-shot services retain their previous behavior. No caller regression was found. + +### Validation and limits + +Installer suite: 724 tests passed, 13 skipped, 80.443 seconds on Python 3.15.2. Ran unittest discovery with PYTHONPATH=installer and a subprocess guard excluding real Docker invocations, including the Compose config test, under the inherited Docker restriction. + +`tools/docs_check.py` passed: 218 files checked, 8 exempt, 67 warnings. Logs are `/tmp/pr85-r2-installer.log` and `/tmp/pr85-r2-docs.log`. CI snapshot: 10 checks passed; Semgrep and Python 3.11/3.12 unit jobs were pending. Passed checks were docs build/links, manifest policy, gitleaks, Trivy, PowerShell, shell lint, and macOS plus both Ubuntu dry runs. No CI polling loop was used; merge still requires the remaining CI checks to finish successfully. + +Reviewed the updated detached `rev-pr85` worktree and the r1→r2 diff. No source edits, commits, pushes, approval action, or merge. Real Docker commands were not rerun in r2; real-row evidence is in r1. + +--- + + +## 日本語 + +[→ To English](#english) + +### 判定 + +**PASS、R1-01 解消。新たな阻害事項なし。** Status の欠落・空・未知の形式は `ok=False`、`health="uptime unknown"` になる。既知の分・時間・日・週は秒数に変換され、不明と区別される。追加された拒否試験は Status 欠落・空・`Up for a while`・`Created` を含む。成功例は20秒境界、約1分・約1時間、分・時間・日・週を含む。短い稼働時間の拒否、health の接尾辞、明示的な healthy の免除、restarting の拒否も引き続き試験される。SPEC-02 0.41 の英和は共に稼働時間不明時の拒否を説明し、旧版の履歴も残っている。 + +### 呼び出し元の検証 + +accept の `default_health_check`、scan の `services_healthy`、up と秘密の更新が使う `wait_until_healthy` を確認した。独立した runner 注入試験で実際の accept/scan の関数を通し、Status 欠落・空・未知・19秒で拒否、20秒・約1分・約1時間・2日・3週で成功を確認した。別の19秒→20秒の系列でポーリング関数が成功することも確認した。一時的な Compose JSON を使い、Docker や外部 HTTP は呼んでいない。 + +成功した up の直後の一回だけの accept/scan 検査には、新しい待機期間は発生しない。up が既に同じ閾値を満たしているためである。up 以後の再起動、直接 Compose 起動後20秒未満、稼働時間の証拠欠落では正しく拒否する。明示的に healthy なサービスと正常終了した一回限りのサービスの挙動も維持する。呼び出し元の回帰は見つからなかった。 + +### 検証と範囲 + +installer 全体は Python 3.15.2 で724件成功、13件スキップ、80.443秒。PYTHONPATH=installer を設定して unittest discovery を実行し、前回からの Docker 制限に従って subprocess のガードで Compose config 試験を含む実 Docker 呼び出しを除外した。 + +`tools/docs_check.py` は218ファイル、8免除、67警告で成功。ログは `/tmp/pr85-r2-installer.log` と `/tmp/pr85-r2-docs.log`。CI は確認時点で10項目成功、Semgrep と Python 3.11/3.12 の単体試験が実行中だった。成功済みは文書ビルド・リンク、manifest 方針、gitleaks、Trivy、PowerShell、shell lint、macOS と Ubuntu 2種類の dry run。CI のポーリングループは行っていない。マージ前に残りの CI 成功を確認する必要がある。 + +更新した detached の rev-pr85 worktree と r1→r2 差分を確認した。ソース編集・コミット・push・承認操作・マージはしていない。r2 で実 Docker コマンドは再実行せず、実機の行の証拠は r1 にある。 diff --git a/docs/reviews/2026-11-16_RDM-developer-guide_PR86_S3_aipf_OIDC_r1.md b/docs/reviews/2026-11-16_RDM-developer-guide_PR86_S3_aipf_OIDC_r1.md new file mode 100644 index 00000000..a51ea419 --- /dev/null +++ b/docs/reviews/2026-11-16_RDM-developer-guide_PR86_S3_aipf_OIDC_r1.md @@ -0,0 +1,95 @@ +# PR86 S3 aipf OIDC review r1 / S3 aipf OIDC レビュー r1 + +> **Summary:** CHANGES REQUESTED for one reproducible profile-isolation regression. Cryptographic interoperability, hardening, and committed evidence checks passed; upstream form-token logging deserves follow-up. +> **概要:** プロファイル分離の再現可能な退行一件により CHANGES REQUESTED。暗号の相互運用性、強化設定、コミット済み証跡の検査は合格し、上流のフォームトークンのログには追跡対応を勧める。 + +**Jump:** [English](#english) · [日本語](#japanese) + +- Version / 版: 0.1 — 2026-11-16. **Learned:** a test-body profile guard cannot prevent a module fixture from failing first. / 試験本体のプロファイル判定では、それより先に動くモジュール fixture の失敗を防げない。 +- Applies to: RCOSDP/RDM-developer-guide PR #86, head `853053ede6b6d9d2c0550c2e3c3aa1056c53d370`, base `develop`; REV round 1. +- Related: [Review index](../rev-pr86/docs/reviews/README.md) · [SPEC-06](../rev-pr86/docs/spec/06_documentation_conventions.md) · [SPEC-04](../rev-pr86/docs/spec/04_acceptance_tests.md) · [ADR-15](../rev-pr86/docs/decisions/ADR-15_Mock_OIDC_Issuer_for_aipf.md). +- Used by: PR #86 revision and coordinator review decision. + +--- + + +## English + +[→ 日本語版へ](#japanese) + +### Verdict and findings + +**CHANGES REQUESTED.** R1-01 is a medium-priority correctness fix. R1-02 is a non-blocking security follow-up. + +**R1-01 — Check profile eligibility before loading the signing key.** `installer/acceptance/test_s3_aipf.py:77–85` requires `OIDC_MOCK_SIGNING_KEY` in a module-scoped fixture. Both tests only call `_aipf_or_skip` inside their bodies (lines 198 and 248), after fixture setup. An existing core/minimum environment without the newly introduced key therefore errors instead of skipping S3. The same ordering defeats the documented missing-port skip when the key is also absent. This violates SPEC-04 and the acceptance README's promise that S3 skips outside full, even though core rendering correctly excludes both new services. Reproduced without Docker using the actual imported `signer` fixture and a temporary `.env` containing only `GRDM_PROFILE=core`: pytest reports one setup error at line 84, while the body guard independently skips. Move eligibility into an appropriately scoped prerequisite of `signer`, preserving a hard failure for a missing key only in an eligible full environment. Add regression coverage for core/minimum without the key, full without its port, and full with its port but no key. + +**R1-02 — Track or suppress upstream form-token access logging.** The pinned aipf source at `94e277da694462516c7a8f12024e3aa0ad67dda7` starts `uvicorn.run` without disabling access logs and registers PATs through `/grdm/token?t=...`. `FormSigner` issues a reusable subject-bound token with a 600-second lifetime; the endpoint accepts that token without bearer authentication. A reader of the raw Docker access log can replay it during that window to replace that subject's stored PAT with another valid PAT. It does not directly reveal the user's PAT or vault key. The installer acceptance path does not collect Docker logs, the report command bundles structured reports rather than raw container logs, and the committed run contains no such token. Traces are off by default and quarantined when enabled; pytest failures are sanitized. Thus no public leak was found in this PR's evidence, but raw container logs should not be published and upstream should suppress/redact this route's query string. Generic long-string masking is not a substitute for explicit handling of `t=` in any future log collector. + +### Security and correctness assessment + +- P-256 parameters, affine addition/doubling, modular arithmetic, hash-to-integer usage and ECDSA verification equations are correct on inspection. Private scalars and signing nonces use `secrets.randbelow(N - 1) + 1`, giving uniform rejection-sampled draws in `[1,N-1]`; there is no deterministic nonce reuse or modulo bias. Zero `r` or `s` causes retry. Random collision probability is negligible, not mathematically zero. +- Signatures are JOSE's fixed-width 32-byte `r` followed by 32-byte `s`, not ASN.1 DER. The committed unit test verifies the RFC 7515 A.3 external signature and rejects a modified message. Independently tested 30 generated keys against installed `cryptography` 50.0.2: public points match and signatures verify in both directions, converting DER with `utils` only at the reference boundary. +- Python scalar multiplication and inverses are not constant-time. For this explicitly local development fixture, with no remotely callable signing endpoint, the documented limitation is acceptable; it must not become a production signer. `installer/pyproject.toml` has `dependencies = []`; cryptography is not an installer or acceptance declared dependency. Its presence in aipf and in the reviewer's existing venv does not make it an installer dependency. Prefer audited cryptography if that dependency policy changes, but replacement is not required for this review. +- The image copies only the public-document server and contains no baked-in key or signing implementation. JWKS output selects public members only. There is no arbitrary signing/token endpoint. Runtime private key material does enter the mock container environment: “never leaves runtime/.env” is not literally true, although nothing exposes it through HTTP. Sending only a derived public JWK would reduce exposure further. +- Rendered mock service joins internal `grdm` only, has no host port, runs as 65534, read-only, drops all capabilities, and has no-new-privileges. The aipf service is uid 1000, read-only, drops all capabilities, has no-new-privileges, publishes only on 127.0.0.1, and uses writable tmpfs `/tmp` for the vault. `/healthz` must return 200. Core service selection excludes aipf and oidc_mock. +- Fernet generation uses 32 OS-random bytes with canonical padded URL-safe base64. New key validation runs at init/render, errors identify keys without values, and existing atomic `.env` mode 0600 and directory 0700 logic is unchanged. No PAT/vault-key logging was found in the added installer paths. +- S3 implements the rewritten success, idempotence, signature/audience/actor rejection, forged form, and cross-user write refusal contract. Successful `write_record` remains explicitly out of scope because internal WaterButler URLs are unresolved. English/Japanese S3 contracts agree. SPEC-02 is 0.42, earlier versions remain, and ADR-15 is unique in the decision directory; #86 was the only open PR at inspection. +- Inspected all five files in `acc-20261115T162657Z-3b9bef`: 20 passed, four skipped, no failures; actual hostname is replaced with `redacted-host`; no `/Users/` path, email, JWT, PAT, form token or secret was found. This is review of supplied evidence, not a new live S3 run. + +### Validation and limits + +Commands ran from the isolated `rev-pr86` checkout; no application stack operations, commits, pushes, approvals or merges occurred. + +```toon +checks[6]{check,result}: + installer_unittest,"746 tests; OK; 12 skipped; 97.886 s (system python)" + tools_unittest,"101 tests; OK" + docs_check,"PASS; 219 files; 8 exempt; 70 warnings" + manifest_check,"PASS; 0 findings; 878 policy-scanned files" + cryptography_interop,"30 keys; bidirectional signature and public-point checks passed" + core_fixture_reproduction,"expected regression reproduced: 1 setup error instead of skip" +``` + +CI snapshot: 13 checks passed, including both Python 3.11/3.12 unit jobs, docs build/links, manifest policy, gitleaks, Semgrep, Trivy, shell/PowerShell and OS dry runs. The amd64+arm64 image-build job was pending; it was not polled to completion. No isolated Docker image build/run was needed; no full-stack acceptance rerun was authorized. Existing unit skips and docs warnings are reported, not counted as full coverage. Fix R1-01, rerun its regression and normal suites, and confirm the pending image job before final integration. + +--- + + +## 日本語 + +[→ To English](#english) + +### 判定と指摘 + +**CHANGES REQUESTED。** R1-01 は中優先度の正しさの修正、R1-02 はマージを妨げないセキュリティ上の追跡事項です。 + +**R1-01 — 署名鍵を読む前にプロファイルの対象判定を行う。** `installer/acceptance/test_s3_aipf.py:77–85` の module scope の `signer` fixture は OIDC 鍵を必須とします。両試験の `_aipf_or_skip` は本体(198、248 行)にあり、fixture より後です。新設の鍵を持たない既存 core/minimum 環境では、S3 がスキップでなくエラーになります。ポートと鍵の両方が無い場合も、仕様のポート欠如によるスキップより鍵のエラーが先です。core の描画が二サービスを含まない点は正しいものの、SPEC-04 と acceptance README の「full 以外でスキップ」に反します。Docker を使わず、実際の `signer` を import し、`GRDM_PROFILE=core` だけの一時 `.env` を使う pytest で 84 行の setup error 一件を再現しました。本体の判定だけなら skip します。適切な scope の前提 fixture で対象を先に判定し、実行対象の full で鍵が無い場合だけは引き続き失敗にしてください。鍵なし core/minimum、ポートなし full、ポートあり鍵なし full を回帰試験に含めてください。 + +**R1-02 — 上流のフォームトークンのアクセスログを追跡または抑止する。** 固定した aipf の `94e277da694462516c7a8f12024e3aa0ad67dda7` はアクセスログを無効にせず `uvicorn.run` を起動し、PAT 登録の URL は `/grdm/token?t=...` です。`FormSigner` は subject に結び付く有効期間 600 秒の再利用可能なトークンを発行し、この口は bearer 認証を要求しません。Docker の生ログを読める人は、有効期間内なら別の有効 PAT を送って、その subject に預けた PAT を置き換えられます。元の PAT や vault 鍵を直接読む機能ではありません。installer の受入処理は Docker ログを採取せず、report は生ログでなく構造化された報告をまとめます。今回の証跡にも当該トークンはありません。trace は既定で無効、有効時は隔離され、pytest の失敗出力はマスクされます。今回の公開証跡で漏洩は見つかりませんでしたが、生ログは公開せず、上流でこの経路の query を消すか隠す対応を勧めます。将来ログを採取するなら、一般的な長文字列のマスクだけでなく `t=` を明示的に扱う必要があります。 + +### セキュリティと正しさ + +- P-256 の定数、点の加算・倍算、剰余計算、ハッシュの整数化、検証式は確認した範囲で正しい。秘密スカラーと nonce は `secrets.randbelow(N - 1) + 1` による一様な拒否サンプリングであり、意図的な再利用や剰余による偏りはない。`r` または `s` がゼロなら再試行する。ランダム衝突は無視できる確率だが数学的にゼロではない。 +- 署名は DER でなく、各 32 バイトの `r || s` という JOSE 形式。既存試験は RFC 7515 A.3 の外部署名ベクトルと改変拒否を確認する。独立に既存 venv の cryptography 50.0.2 で 30 鍵を確かめ、公開点と双方向の署名検証が一致した。DER 変換は比較相手の境界だけに使用した。 +- Python の点の乗算と逆数は定時間ではない。署名 API を公開しない明示的なローカル開発 fixture として、この文書化された制約は許容できるが、本番署名器には使わない。installer の `dependencies = []` であり、cryptography は installer/accept の宣言依存ではない。aipf やレビュワの venv に入っていることとは別問題。依存方針が変わるなら監査済み実装を優先したいが、今回の必須修正ではない。 +- イメージに入るのは文書配布サーバのみで、鍵の埋め込みも署名実装もない。JWKS は公開メンバーのみを選ぶ。任意署名や token endpoint はない。ただし秘密鍵は実行時に mock コンテナの環境変数へ渡るため、「runtime/.env の外へ出ない」は文字どおりには正しくない。HTTP には出ない。公開 JWK だけを渡す設計ならさらに露出を減らせる。 +- mock は internal な `grdm` のみ、ホストポートなし、uid 65534、read-only、全 capability 削除、no-new-privileges。aipf は uid 1000、同じ強化設定、127.0.0.1 のみ公開し、vault は tmpfs `/tmp` に置く。health は `/healthz` の 200 を要求する。core のサービス選択には両サービスを含まない。 +- Fernet 鍵は OS の乱数 32 バイトから正規の padding 付き URL-safe base64 を作る。init/render の形式検査は値でなくキー名を示す。`.env` の原子的書込みと 0600、親ディレクトリ 0700 は維持されている。追加 installer 経路に PAT/vault 鍵のログ出力は見つからない。 +- S3 の成功、二度目の同一結果、署名・宛先・代理者の拒否、偽フォーム、他人のプロジェクトへの書込み拒否は改訂仕様に対応する。成功側の `write_record` は内部 WaterButler URL が未解決として明示的に対象外。S3 の英和は一致し、SPEC-02 は 0.42 で旧版を保持。ADR-15 は決定ディレクトリで一意で、確認時の open PR は #86 だけだった。 +- 実行フォルダ `acc-20261115T162657Z-3b9bef` の五ファイルを確認。20 合格、4 skip、失敗なし。実ホスト名は `redacted-host` に置換され、`/Users/`、メール、JWT、PAT、フォームトークン、秘密値は見つからない。提出済み証跡の確認であり、新しい実機 S3 実行ではない。 + +### 検証と制限 + +独立した `rev-pr86` で実行し、スタックの操作、commit、push、approve、merge は行っていません。 + +```toon +checks[6]{check,result}: + installer_unittest,"746 件; OK; 12 skip; 97.886 秒(system python)" + tools_unittest,"101 件; OK" + docs_check,"PASS; 219 ファイル; 8 除外; 70 警告" + manifest_check,"PASS; 指摘 0; ポリシー検査 878 ファイル" + cryptography_interop,"30 鍵; 双方向署名と公開点の一致" + core_fixture_reproduction,"退行を再現: skip の代わりに setup error 一件" +``` + +CI は確認時点で 13 件合格。Python 3.11/3.12 の単体試験、文書ビルド・リンク、manifest、gitleaks、Semgrep、Trivy、shell/PowerShell、各 OS の dry run を含みます。amd64+arm64 のイメージビルド一件は pending で、完了まで繰り返し待っていません。単独 Docker ビルド・実行は不要と判断し、スタック全体の受入再実行は許可範囲外のため行っていません。既存 skip と文書警告を完全な網羅とは扱いません。R1-01 の修正、回帰試験と通常スイートの再検査、および残る CI の確認が必要です。 diff --git a/docs/reviews/2026-11-16_RDM-developer-guide_PR86_S3_aipf_OIDC_r2.md b/docs/reviews/2026-11-16_RDM-developer-guide_PR86_S3_aipf_OIDC_r2.md new file mode 100644 index 00000000..52d4de1a --- /dev/null +++ b/docs/reviews/2026-11-16_RDM-developer-guide_PR86_S3_aipf_OIDC_r2.md @@ -0,0 +1,87 @@ +# PR86 S3 aipf OIDC review r2 / S3 aipf OIDC レビュー r2 + +> **Summary:** PASS: R1-01 is fixed and the mock receives only derived public key material. The independent upstream access-log review remains a separate CHANGES REQUESTED. +> **概要:** R1-01 は修正され、mock に渡るのは導出した公開鍵だけとなったため PASS。別途の上流アクセスログレビューは CHANGES REQUESTED のまま。 + +**Jump:** [English](#english) · [日本語](#japanese) + +- Version / 版: 0.1 — 2026-11-16. **Learned:** a fixture dependency enforces the eligibility check before key loading, and derived public settings remove the issuer's need to receive a signing key. / fixture の依存で鍵の読込みより先に対象判定を保証でき、公開設定の導出で発行者へ署名鍵を渡す必要も無くなる。 +- Applies to: RCOSDP/RDM-developer-guide PR #86, head `6fccbd1`, base develop; REV round 2. +- Related: [Review index](../rev-pr86-r2/docs/reviews/README.md) · [SPEC-06](../rev-pr86-r2/docs/spec/06_documentation_conventions.md) · [ADR-15](../rev-pr86-r2/docs/decisions/ADR-15_Mock_OIDC_Issuer_for_aipf.md) · [r1](2026-11-16_RDM-developer-guide_PR86_S3_aipf_OIDC_r1.md) · [Upstream log review](2026-11-16_aipf_PR1_AccessLog_r1.md). +- Used by: PR #86 integration decision. + +--- + + +## English + +[→ 日本語版へ](#japanese) + +### Verdict and prior findings + +**PASS. No new blocking findings.** R1-01 is resolved: session-scoped `signer(aipf_url)` explicitly depends on the eligibility fixture, and both S3 test signatures request eligibility/signing before browser fixtures. Actual pytest runs without an installed browser correctly skip core, minimum with a stale port, and full without the port; eligible full without its key errors on the key before browser startup. Seven fixture tests passed, including the real pytest subprocess cases. Temporarily substituting the r1 S3 file caused the same seven-test suite to fail (five failure and five error records, including subtests); actual pytest cases failed too, so this is not merely a source-shape assertion. Restored HEAD and reran all seven successfully; worktree is clean. + +R1-02 was a non-blocking upstream logging concern. Its attempted fix is reviewed separately in aipf PR #1: encoded parameter names still leak, so that PR receives CHANGES REQUESTED. This does not reopen the guide's resolved R1-01 or establish a leak in its committed evidence. + +### Public-key handling and merge resolution + +`derived_values` validates the private P-256 JWK with `decode_key`, then encodes only `public_jwk`'s explicit public-member allowlist. Init, missing-key repair and rotation refresh the derived setting; render replaces an absent or stale public value after validating secrets. Tests cover these paths and ensure the rendered services contain neither the private-key environment reference nor `d`. The mock template receives only `OIDC_MOCK_PUBLIC_JWK`; the image rejects a supplied `d` member and emits no private value in the error. Its JWKS matches the signing key's public point and verifies the test JWT. The private key remains in the host `.env` for S3. Tag 2 and ADR-15 0.2 reflect the new interface; no signing endpoint or host publication was added. + +The #87 merge retains one `fernet_key()` implementation, used for both RDMCP_CREDENTIAL_ENCRYPTION_KEY and GRDM_VAULT_KEY. Canonical 32-byte padded URL-safe base64 validation remains, alongside CAS, OAuth-column and OIDC-key checks. RDMCP OAuth formats and aipf signing generation coexist in the generator table. The health service list, paths and exact-200 set contain both rdmcp and aipf. The English and Japanese port tables both contain 12 identical rows, including CAS 8090, rdmcp 8100 and aipf 8768; the mock has no published row. SPEC-02 0.43, SPEC-04 0.16, board 0.30 and acceptance README 0.8 retain the merged earlier history. Tests retain core exclusion of both aipf and mock. + +All five files of `acc-20261115T172743Z-b28b73` were inspected. They report S3-only success and rejection: two passed, zero failed/errors/skips. Hostname is `redacted-host`; no actual host identity, `/Users/` path, email, JWT, form token or secret value was found. This is supplied evidence; no live stack was operated during review. + +### Validation and limits + +Fresh detached worktree under the requested scratchpad; no commit, push, approval, merge or stack operation. + +```toon +checks[7]{check,result}: + installer_clean_python,"773 tests; OK; 14 skipped; 98.357s" + s3_fixture_accept_venv,"7 tests; OK; includes real pytest subprocesses" + r1_mutation,"regression detected; five failure and five error records; then restored" + tools,"101 tests; OK" + docs_check,"PASS; 219 files; 8 exempt; 70 warnings" + manifest_check,"PASS; zero findings; 894 policy-scanned files" + CI_snapshot,"13 passed; amd64+arm64 image build pending" +``` + +An initial full run using the existing acceptance/scan venv reported 20 failures and one error, all in unchanged scan tests: `_exe('semgrep')` found the venv's absolute executable, while the fake runner recognizes only literal `semgrep`, returning rc 127/no JSON. Reproduced that cause separately; `scan.py` and `test_scan.py` have no diff from r1. The normal system-Python suite passes, and acceptance fixture tests were independently run in the dependency-equipped venv so they did not hide behind skips. This environment-sensitive existing test limitation is disclosed rather than counted as a product regression. No long CI polling; confirm the remaining image job before integration. + +--- + + +## 日本語 + +[→ To English](#english) + +### 判定と前回指摘 + +**PASS。新しい阻害指摘はない。** R1-01 は解決。session scope の `signer(aipf_url)` が対象判定に明示的に依存し、両試験はブラウザ fixture より先に対象判定と署名を要求する。ブラウザ未導入の実 pytest で、core、古いポートの残る minimum、ポートなし full が skip し、対象の full で鍵が無い場合はブラウザ起動前に鍵のエラーになる。実 pytest subprocess を含む七試験が合格した。S3 ファイルだけを r1 に一時的に戻すと、同じ七試験で failure 5、error 5(subtest 含む)を検出し、実 pytest ケースも失敗したため、ソースの形だけを見た検査ではない。HEAD を復元して七試験を再実行し合格、worktree は clean。 + +R1-02 は上流のログに関する非阻害の追跡事項。その対応案を aipf PR #1 として別途確認したが、エンコードされた名前で漏れるため、その PR は CHANGES REQUESTED。guide の R1-01 を再度未解決にするものでも、guide の証跡に漏洩を発見したものでもない。 + +### 公開鍵とマージの解決 + +`derived_values` は `decode_key` で秘密 P-256 JWK を検証し、`public_jwk` の明示的な公開メンバーのみをエンコードする。init、missing、rotate は導出値を更新し、render も秘密値の検証後に公開値の欠如・陳腐化を修復する。これらを試験し、描画したサービスに秘密鍵の環境変数参照も `d` の値も無いことを確認する。mock のテンプレートが受け取る鍵は `OIDC_MOCK_PUBLIC_JWK` だけ。イメージは `d` を持つ値を拒否し、エラーに秘密値を出さない。JWKS は署名鍵の公開点と一致し試験の JWT を検証できる。秘密鍵は S3 用にホストの `.env` に残る。tag 2 と ADR-15 0.2 が新しいインターフェースを反映し、署名 endpoint やホスト公開は増えていない。 + +#87 のマージ後も `fernet_key()` は一実装で、RDMCP_CREDENTIAL_ENCRYPTION_KEY と GRDM_VAULT_KEY が共有する。32 バイト、padding 付き URL-safe base64 の正規形検査を維持し、CAS・OAuth カラム長・OIDC 鍵の検査と共存する。生成器には RDMCP OAuth と aipf 署名鍵の双方が残る。health の対象・パス・200 必須の集合には rdmcp と aipf の両方がある。英和のポート表は一致する 12 行で、CAS 8090、rdmcp 8100、aipf 8768 を含み、mock の公開行は無い。SPEC-02 0.43、SPEC-04 0.16、board 0.30、acceptance README 0.8 は統合された過去の履歴を保持する。core から aipf と mock を除く試験も残る。 + +`acc-20261115T172743Z-b28b73` の五ファイルを確認。S3 だけの成功・拒否が 2 passed、失敗・error・skip はゼロ。ホスト名は `redacted-host` に置換され、実ホスト情報、`/Users/`、メール、JWT、フォームトークン、秘密値は見つからない。提出済み証跡の確認であり、レビュー中に実スタックは操作していない。 + +### 検証と制限 + +指定 scratchpad の新規 detached worktree を使用。commit、push、approve、merge、スタック操作なし。 + +```toon +checks[7]{check,result}: + installer_clean_python,"773 件; OK; 14 skip; 98.357 秒" + s3_fixture_accept_venv,"実 pytest subprocess を含む 7 件; OK" + r1_mutation,"退行を検出; failure 5 と error 5; 復元済み" + tools,"101 件; OK" + docs_check,"PASS; 219 ファイル; 8 除外; 70 警告" + manifest_check,"PASS; 指摘ゼロ; ポリシー検査 894 ファイル" + CI_snapshot,"13 成功; amd64+arm64 イメージビルドは pending" +``` + +最初に既存の accept/scan venv で全体を動かした際は、変更のない scan 試験で failure 20、error 1 となった。`_exe('semgrep')` が venv 内の絶対パスを返す一方、fake runner は literal `semgrep` だけを認識し、rc 127/JSON なしを返すことを別途再現した。`scan.py` と `test_scan.py` は r1 から差分なし。通常の system Python では全体が合格し、依存の必要な受入 fixture 試験は別途 venv で実行して skip に隠れないことを確認した。既存試験の環境依存の制約として記録し、製品の退行とは扱わない。CI は長時間 polling せず、統合前に残るイメージジョブを確認する必要がある。 diff --git a/docs/reviews/2026-11-16_RDM-developer-guide_PR87_S2_rdmcp_r1.md b/docs/reviews/2026-11-16_RDM-developer-guide_PR87_S2_rdmcp_r1.md new file mode 100644 index 00000000..06eb1b5a --- /dev/null +++ b/docs/reviews/2026-11-16_RDM-developer-guide_PR87_S2_rdmcp_r1.md @@ -0,0 +1,114 @@ +# REV r1 S2 real CAS OAuth review / S2 本物の CAS OAuth レビュー + +> **Summary:** PR A has no blocking regression; PR B is technically acceptable but must wait for the merged rdmcp pin. The Linux MIME failure also occurs on develop. +> **概要:** PR A に妨げとなる回帰はない。PR B の実装は妥当だが、rdmcp のマージ後の固定版を待つ。Linux の MIME 失敗は develop でも起きる。 + +**Jump:** [English](#english) · [日本語](#japanese) + +- Version / 版: 0.1 — 2026-11-16. **Learned:** CAS uses whole-string service matching, and a real CAS bearer validator does not replace fakecas browser authentication. / CAS のサービス照合は文字列全体の一致であり、本物の CAS による Bearer 検証は fakecas のブラウザ認証を置き換えない。 +- Applies to: RCOSDP/RDM-rdmcp-dev2027 PR #1 `9c6e0050468451bba0c0cb9e3f8fdbb45f0b310b` (A); RCOSDP/RDM-developer-guide PR #87 `e05f6205c5de382e9a00f3cf0945f4eac58a05bd` (B). +- Reviewer: REV(担当レビュワ, Codex Astra) +- Related: [Review index](../rev-pr87/docs/reviews/README.md), [SPEC-06](../rev-pr87/docs/spec/06_documentation_conventions.md), [SPEC-04](../rev-pr87/docs/spec/04_acceptance_tests.md), [SPEC-02](../rev-pr87/docs/spec/02_installer_cli.md). +- Used by: coordinator's round-1 merge decision; this record is uncommitted. + +--- + + +## English + +[→ 日本語版へ](#japanese) + +### Verdicts + +**A: PASS with one nonblocking configuration-validation nit.** No P0/P1/P2 regression found. **B: HOLD for the explicitly documented dependency gate; implementation PASS.** Do not merge the current PR-head pin. Merge A first, update B to A's merge commit, and verify the final diff and manifest checks. No merge, approval, push, commit, or grdm-dev stack operation was performed. + +### Findings and security assessment + +**A-01 — P3, optional hardening (`src/rdmcp/config.py:74–81`).** The validator tests truthiness of username/password and never reads `url.port`. Reproduced acceptance of `http://@127.0.0.1`, `http://localhost:invalid`, and `http://127.0.0.1:99999`. Reject userinfo by testing `is not None` and validate the port. Add cases for empty userinfo, malformed ports, IPv6 and hostname tricks. This is a configuration-contract issue, not an off-loopback bypass: malicious suffix hosts, both directions of nonempty userinfo, IPv6 loopback/mapped IPv6 and a backslash/userinfo trick were rejected. Normal localhost and 127.0.0.1 are the only HTTP hosts accepted. + +Issuer, resource-server origin and RDM callback derive from the same `public_url`; the callback appends `/oauth/callback`. Defaults preserve the prior HTTPS server-name behavior. `token_base_url` reaches only the server's HTTP POST for code exchange and refresh; authorization still uses the browser-facing CAS URL. PKCE, state persistence/consumption and grants are unchanged. `/healthz` returns only literal `ok`, without credentials or upstream calls. Its HTTP test passes. + +**B-01 — mandatory merge gate (`manifest/upstreams.json:700`).** The PR body explicitly says not to merge before A is merged and the pin moves to its merge commit. Current `9c6e0050…` is the reviewed PR head, not that merge commit. This hold is already acknowledged by the author. + +Independent renders of minimum, core and full confirmed that only full publishes CAS (8090), includes rdmcp (8100) and overrides api's `CAS_INTERNAL_SERVER_URL`. All published ports bind to 127.0.0.1. Existing core/minimum runtime secrets need the documented `secrets init --missing` migration because the required key list grows globally; runtime service isolation is otherwise maintained. Tests cover full-to-core callback-file removal and moved ports. + +The OAuth secret and Fernet key are generated randomly, passed by environment reference and not interpolated into the Python seed source. The ORM seed finds the existing named application and updates it on subsequent runs, so ordinary serial startup is idempotent; it has no shell/Python/SQL injection from the supplied values. Its callback is exactly the generated public rdmcp origin plus `/oauth/callback`, and only that callback is printed. The added acceptance password enters existing seed-secret masking. No real OAuth secret was found in tracked changes or either committed run folder. + +CAS callback matching is **exact, not an open-redirect regex**. `OAuthCallbackAuthorizeService` inherits `RegexRegisteredService.matches`, which uses Java `matcher(...).matches()` (whole input). Dots are escaped and the port/path are literal. A terminal `$` is unnecessary and would violate the subclass's requirement that the configured string end in `/oauth2/callbackAuthorize`. The OSF OAuth client's redirect is separately checked with string equality (case-insensitive) by CAS. Sources inspected: pinned cas-overlay `22610b7681a2`, and Apereo CAS v4.1.5 `OAuthCallbackAuthorizeService` / `RegexRegisteredService` through GitHub API. + +The api override selects a real validator rather than bypassing authentication or authorization. Pinned CAS supports personal tokens as well as issued OAuth tokens; pinned osf.io `07cc56d47b39` preserves user checks and scopes. **Fakecas browser-session users remain authenticated by the OSF cookie path** (`OSFSessionAuthentication`, before `OSFCASAuthentication`), which reads the shared session database without contacting CAS. Therefore this setup does not turn the whole full profile into password-verified CAS login: a fakecas user can still act through their session and create/use PATs. Fakecas's synthetic GUID accessToken is not a real CAS OAuth token; clients depending on it as a standalone bearer must not assume it works. Cookie/PAT success does not establish that compatibility. Real-CAS tokens sent to web/v1 still fail against fakecas, as the PR documents. These are limits of the explicitly chosen split, not evidence of stronger end-to-end login assurance; retain this distinction in follow-up authentication work. + +S2 implements the specified success flow and all rejection checks: anonymous 401 plus Bearer challenge, invisible user_b project and forged selection 400, denied tool calls for user_b and the user's own unselected project, and a subsequent successful granted call. The own-unselected project is independently readable before the grant test, so the denial is meaningful. CAS form requests use the guarded HTTP wrapper; node selection uses the guarded browser. This checks forms, not CAS JavaScript rendering. SPEC-04 EN/JA agree and prior version entries remain in SPEC-02/04, status and acceptance README. + +Both run folders contain five text files each. Inspection and pattern checks found no actual hostname, `/Users` path, email, bearer token or secret; XML hostname is `redacted-host`. Their evidence reports S2 2 passed and whole-suite 20 passed / 4 skipped. These are author-provided live-run records, not an independently repeated acceptance run. The documented temporary exclusion of aipf_mcp_data_grdm means they do not prove every unmodified full-profile service was healthy. + +### Verification + +Results from fresh scratch clones; the installer uses Python 3.15.2, and rdmcp uses Python 3.12.16 locally. The Linux run used a disposable existing rdmcp image, with reviewed source and tests transferred over stdin, no stack volumes or networks attached. + +```toon +checks[8]{check,result}: + installer unittest,"739 run; OK; 12 skipped" + tools unittest,"101 run; OK" + docs_check,"PASS; 218 checked; 8 exempt; 67 warnings" + manifest_check,"PASS; policy scan 876 text files; 0 findings" + manifest toon --check,PASS + rdmcp pytest macOS,"122 passed" + rdmcp pytest Linux,"121 passed; 1 failed: test_detect_mimetype_infers_office_from_materialized_path" + same Linux test on develop 7eed0db,"same failure: text/plain instead of presentation MIME" +``` + +The failing test and MIME implementation are unchanged by A. The Linux baseline comparison confirms the reported pre-existing failure; macOS has the needed MIME mapping and passes. `tools/scan_secrets.py --all` in A found no actual values. Full Linux log: [rdmcp-linux-pytest.log](rdmcp-linux-pytest.log). Other logs: [installer](installer-tests.log), [tools](tools-tests.log), [docs](docs-check.log), [manifest](manifest-check.log), [macOS pytest](rdmcp-pytest.log). + +Both `gh pr view ... --json statusCheckRollup` and `gh run list --commit ` returned empty arrays for A and B. No completed or running CI was visible at review time; local checks must not be described as green CI. Both PRs were read using `gh pr diff`. No long polling was used. + +--- + + +## 日本語 + +[→ To English](#english) + +### 判定 + +**A: PASS(設定検証の非必須改善1件)。** P0/P1/P2 の回帰は見つからなかった。**B: 明記済みの依存マージ条件により HOLD、実装は PASS。** 現在の PR 先頭への固定のままマージしない。A をマージして B の固定版をそのマージコミットへ移し、最終差分と manifest 検査を確認する。マージ・承認・push・commit・grdm-dev スタック操作は行っていない。 + +### 指摘と安全性 + +**A-01 — P3、任意の改善(`src/rdmcp/config.py:74–81`)。** username/password を真偽値で調べ、`url.port` を読まないため、`http://@127.0.0.1`、`http://localhost:invalid`、`http://127.0.0.1:99999` を受け付けることを再現した。userinfo は `is not None` で拒み、ポートも検証するとよい。空 userinfo、不正ポート、IPv6、偽装ホストのテストを足す。設定契約の問題であり、外部ホストへの抜け道ではない。悪意ある接尾ホスト、空でない userinfo の両方向、IPv6 と IPv4 射影 IPv6、バックスラッシュと userinfo の組合せは拒否された。HTTP の許容ホストは localhost と 127.0.0.1 に限られる。 + +issuer・resource server の起点・RDM callback は同じ `public_url` から導出され、callback は `/oauth/callback` を付ける。未設定時は従来の HTTPS server-name 動作を保つ。内部 CAS URL はサーバーのコード交換・refresh POST だけに使い、認可 URL は公開 CAS のまま。PKCE・state の保存と消費・grant は変更されていない。`/healthz` は資格情報も上流アクセスもなく、固定文字列 `ok` だけを返し、HTTP テストも通る。 + +**B-01 — 必須のマージ条件(`manifest/upstreams.json:700`)。** PR 本文は A のマージとマージコミットへの固定変更までマージしないと明記する。現在の `9c6e0050…` はレビュー対象の PR 先頭であり、そのマージコミットではない。著者も把握済みの保留である。 + +minimum・core・full を独立に描画し、CAS 公開(8090)・rdmcp(8100)・api の `CAS_INTERNAL_SERVER_URL` 上書きが full だけであることを確認した。公開ポートはすべて 127.0.0.1 に束縛される。必須鍵の一覧は全体で増えるため、既存 core/minimum の秘密ファイルには記載済みの `secrets init --missing` が必要だが、サービス分離は維持する。full から core に戻す callback ファイル削除とポート移動にもテストがある。 + +OAuth secret と Fernet 鍵は無作為に生成し、環境変数参照で渡す。seed の Python ソースに値を埋め込まない。ORM は既存の同名アプリを見つけて更新するため、通常の直列起動は冪等で、値から shell/Python/SQL を注入する道はない。callback は生成した rdmcp 公開起点に `/oauth/callback` を付けた厳密な値で、出力も callback だけ。追加した受入用パスワードは既存の seed-secret マスクに入る。追跡変更や二つの実行フォルダーに実 OAuth secret は見つからなかった。 + +CAS callback は **完全一致で、open redirect を許す広い正規表現ではない**。`OAuthCallbackAuthorizeService` が継承する `RegexRegisteredService.matches` は Java の `matcher(...).matches()` で入力全体を照合する。ドットはエスケープし、ポートとパスは固定文字列。末尾 `$` は不要であり、追加すると設定文字列が `/oauth2/callbackAuthorize` で終わるという派生クラスの条件に反する。OSF OAuth クライアントの redirect は別途 CAS が文字列一致(大小文字を区別しない)で調べる。固定 cas-overlay `22610b7681a2` と、GitHub API 経由の Apereo CAS v4.1.5 の両クラスを確認した。 + +api の上書きは実際の検証先を選ぶもので、認証・認可を迂回しない。固定 CAS は PAT と自身が発行した OAuth token の両方を扱い、固定 osf.io `07cc56d47b39` のユーザー検査・scope も残る。**fakecas のブラウザセッション利用者は OSF cookie 経路で認証され続ける**。`OSFCASAuthentication` より先の `OSFSessionAuthentication` が共有セッションDBを読み、CAS へ問い合わせない。従って full 全体がパスワード確認済み CAS ログインに変わるわけではなく、fakecas 利用者はセッションで操作し PAT を作成・利用できる。fakecas の擬似 GUID accessToken は本物の CAS OAuth token ではなく、それ単体を Bearer とするクライアントの互換性を仮定してはいけない。cookie/PAT の成功はその互換性の証明にはならない。本物の CAS token を web/v1 に送ると fakecas が拒む点は PR に記載済み。この分離方式の限界であり、ログイン全体の保証が強くなった証拠ではない。今後の認証作業でも区別を保つ。 + +S2 は仕様の成功経路と全拒否条件を実装する。匿名 401 と Bearer challenge、user_b の非表示と偽造選択 400、user_b と自分の未選択プロジェクトへのツール拒否、その後の許可済み呼出成功を調べる。自分の未選択プロジェクトは grant 検査前に読めると独立に確かめているので、拒否の意味がある。CAS フォームは守りのある HTTP wrapper、ノード選択は守りのあるブラウザを使う。CAS の JavaScript 描画を試すものではない。SPEC-04 の英和は対応し、SPEC-02/04・進捗・受入 README の旧版も残る。 + +二つの実行フォルダーは各5個のテキストファイル。目視とパターン検査で実ホスト名・`/Users` パス・メール・Bearer token・secret は見つからず、XML hostname は `redacted-host`。S2 2件合格、全体20件合格・4件スキップという証跡である。著者の実機記録を確認したもので、受入試験を独立に再実行してはいない。記載された aipf_mcp_data_grdm の一時除外により、未変更 full 全サービスの健全性を証明する記録ではない。 + +### 検証 + +新しい scratch clone で実行した。installer は Python 3.15.2、rdmcp のローカルは Python 3.12.16。Linux は既存 rdmcp image の使い捨てコンテナへ対象ソースとテストを標準入力で渡し、スタックのボリュームやネットワークには接続していない。 + +```toon +checks[8]{check,result}: + installer unittest,"739件実行・OK・12件スキップ" + tools unittest,"101件実行・OK" + docs_check,"PASS・218件検査・8件除外・67警告" + manifest_check,"PASS・policy scan 876テキスト・検出0" + manifest toon --check,PASS + rdmcp pytest macOS,"122件合格" + rdmcp pytest Linux,"121件合格・1件失敗: test_detect_mimetype_infers_office_from_materialized_path" + develop 7eed0db の同じ Linux テスト,"同じ失敗: presentation MIME の代わりに text/plain" +``` + +失敗したテストと MIME 実装は A で変更されていない。Linux の develop 比較により既存失敗との説明を確認した。macOS は必要な MIME 対応を持ち合格する。A の `tools/scan_secrets.py --all` は実値を検出しなかった。ログは [Linux pytest](rdmcp-linux-pytest.log)、[installer](installer-tests.log)、[tools](tools-tests.log)、[docs](docs-check.log)、[manifest](manifest-check.log)、[macOS pytest](rdmcp-pytest.log)。 + +A・B とも `gh pr view ... --json statusCheckRollup` と `gh run list --commit <対象SHA>` は空配列だった。確認時に完了・実行中 CI は表示されず、ローカル検査を CI 合格とは呼べない。両PRを `gh pr diff` で読み、長時間のポーリングはしていない。 diff --git a/docs/reviews/2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r1.md b/docs/reviews/2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r1.md new file mode 100644 index 00000000..00a2430d --- /dev/null +++ b/docs/reviews/2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r1.md @@ -0,0 +1,123 @@ +# REV PR88 S5 Shibboleth review r1 / S5 Shibboleth レビュー r1 + +> **Summary:** Request changes: removing CAS's host port does not authenticate the sender of institution identity headers. The head.js exception also needs the owner's decision. +> **概要:** 修正要求。CAS のホストポートを消すだけでは機関認証ヘッダの送信元を認証できない。head.js の例外には所有者の判断も必要。 + +**Jump:** [English](#english) · [日本語](#japanese) + +- Version / 版: 0.1 — 2026-11-16. **Learned:** a private Docker network prevents external exposure but does not prove that an identity header came from the SP. / Docker の内部ネットワークは外部公開を防ぐが、認証ヘッダが SP から来た証明にはならないと分かった。 +- Applies to: RCOSDP/RDM-developer-guide PR #88, head `38c64143abd1ce367276ef5f53b52b3685457676`, base `develop`. +- Reviewer: REV(担当レビュワ, Codex Astra). +- Related: [Review index](../rev-pr88/docs/reviews/README.md) · [SPEC-06](../rev-pr88/docs/spec/06_documentation_conventions.md) · [SPEC-04](../rev-pr88/docs/spec/04_acceptance_tests.md) · [ADR-05](../rev-pr88/docs/decisions/ADR-05_Test_IdP_for_Shibboleth.md). Used by: the coordinator's revision and merge decision. + +--- + + +## English + +[→ 日本語版へ](#japanese) + +### Verdict and scope + +**Request changes.** R1-01 is a security blocker. R1-02 requires an owner decision, which this review does not make. Reviewed `gh pr diff 88` and a fresh detached `rev-pr88` worktree at the requested head. GitHub reports `CONFLICTING`; incorporating develop/S3 belongs to the coordinator. No merge, approval, push, commit, or grdm-dev stack operation was performed. No container build/run or live exploit was performed; the trust-boundary finding is based on the actual rendered topology and pinned source, not a claimed live login reproduction. + +### R1-01 — High: CAS accepts unauthenticated institution identity headers from non-SP callers + +Locations: `installer/grdm_dev/templates/cas.json:10`, `installer/grdm_dev/compose.py:683-687`, `installer/grdm_dev/templates/idp/sp-httpd.conf.tmpl:21-22`. The pinned `RDM-cas-overlay-dev2027` source at `22610b7681a29241365108e5129bcd43ade4610e` was read from the local source checkout without changing it: `cas-server-support-osf/src/main/java/io/cos/cas/authentication/handler/support/OpenScienceFrameworkPrincipalFromRequestRemoteUserNonInteractiveCredentialsAction.java:347-414`. + +The action treats any nonempty `AUTH-Shib-Session-ID` as SAML authentication, copies `AUTH-*` values from the request, and sends the normalized identity to the API using CAS's own signing/encryption secrets. It explicitly assumes Apache removed forged headers. There is no sender authentication or mode switch at this decision. Missing `REMOTE_USER` only logs a warning; it does not reject. The login webflow invokes this action before interactive password authentication. + +An isolated render confirms that shibboleth CAS has no published port, but remains on `grdm`, shared by web, api, rdmcp, the IdP and other peers. Any such peer can send forged session, IdP, mail and display-name headers directly to `cas:8080/login`, bypassing both SAML verification and the SP's affiliation rule. The stylesheet accepts the configured test IdP identifier and does not independently verify membership. Removing a host port is necessary but insufficient. + +In full fakecas mode the same action is reachable at `127.0.0.1:8090`, with no SP at all. The default image stylesheet is still mounted implicitly from the image; it maps known IdPs, including `bt`, and `osf_init` populates that institution in the test set. Thus a local client can enter the institution-authentication path using forged headers; eventual account creation/login is subject to downstream institution/account checks, not proof of SAML authentication. This exposed fakecas path predates this PR; it is explicitly reported because the requested review includes it. The newly wired shibboleth path does not repair it. A fresh fakecas render does not use the new mock stylesheet, so merely forging the mock IdP identifier is not a valid reproduction for that mode. + +Inspected web/API CAS client code creates ticket-validation requests and explicit Bearer-token requests, without forwarding arbitrary inbound headers. No generic web/API-to-CAS reverse proxy was found in the inspected configuration. This does not remove direct peer access. + +**Required mitigation:** enforce the trust boundary in CAS before constructing a remote credential. For example, generate a dedicated random SP-to-CAS secret; have the SP overwrite any client-provided copy and set it on its backend request; have CAS compare it and fail closed before accepting any identity header. Keep it out of committed/rendered secret-bearing files by runtime injection. Enable this path only for shibboleth; disable/strip institution headers on all fakecas CAS ingress. An isolated authenticated proxy/network design is an alternative, but every direct backend route must be covered, including ticket-validation callers. Merely clearing headers on the SP cannot protect requests that bypass the SP. + +Add regression tests for forged headers through the SP, direct CAS requests from an ordinary network peer, full fakecas port 8090, missing/wrong trust secret, and valid SP forwarding. Assert no authenticated session/ticket is created for forged requests. Existing S5 success/affiliation-rejection tests do not exercise this boundary. + +### R1-02 — Owner decision: external-asset exception for head.js + +Location: `installer/acceptance/netguard.py:46`, with matching SPEC-04 and status-board text. The new cdnjs `/ajax/libs/headjs//head.min.js` entry makes this request a counted, blocked known asset rather than an unexpected external-request failure. It does **not** permit network egress, but it does relax the acceptance gate. Obtain and record the owner's explicit approval, or remove the exception and serve head.js locally. This reviewer neither approves nor rejects the policy exception. The M4 local-serving follow-up is already recorded. + +### Other security and regression observations + +- The SP uses `ShibUseHeaders On`, clears incoming `REMOTE_USER`/`Remote-User` early, and does not disable spoof checking. Shibboleth documents `checkSpoofing=true` by default; controlled incoming headers are rejected. This is protection for requests traversing the SP, not arbitrary CAS callers. It is not a blanket claim that every possible `AUTH-*` name is stripped. Source: [Shibboleth SP3 InProcess](https://shibboleth.atlassian.net/wiki/spaces/SP3/pages/2063696243/InProcess). +- The RequestMap expression permits anonymous requests and authenticated `member` sessions, and rejects authenticated nonmembers. The declared canonical host/port matches the rendered SP origin. Existing acceptance tests check SP 403, absence of an OSF cookie and API 401 for rejection; success checks the display name and institution through the API. +- Metadata embeds generated public certificates; keys come from `secrets init`, with separate pairs, directory 0700 and keys 0600. IdP salt/admin password are environment references, validated at startup and in PHP; no fixed admin password is present. Mock user passwords are intentional fixtures. The IdP gets only its own private key. +- The IdP base is digest-pinned, and the SimpleSAMLphp 2.5.4 tarball is checked against a fixed SHA-256. Rendered service runs as 1000:1000, read-only, with explicit tmpfs, dropped capabilities and no-new-privileges; host publishing is loopback-only. Build/run was not repeated, so these are configuration findings, not a fresh image-runtime certification. +- CentOS 7 EOL, amd64 emulation, mod_ssl omission and the maintained multi-architecture M4 replacement are explicitly documented in both languages and the status board. +- Changed English/Japanese sections agree on the reviewed design, limits and test results. Earlier version entries remain. Core/minimum retain fakecas selection; new IdP secrets are required only for full shibboleth. Full fakecas still publishes CAS for S2. The documented new secrets/key generation and GRDM_AUTH persistence also apply outside S5; they are not described as byte-for-byte unchanged output. +- All four specified run folders contain text evidence only. Pattern checks and inspection found no real hostname, /Users path, email address, token, SAML assertion or private key. JUnit uses the literal `redacted-host`. The latest S5 run records 2 passed; latest full fakecas run records 20 passed/4 skipped; earlier S5 records 2 passed and earlier full records 18 passed/6 skipped. These are submitted evidence, not newly rerun browser tests; excluded `aipf_mcp_data_grdm` and then-unimplemented S3 are documented. + +### Validation + +Commands ran from `rev-pr88` with Python 3.15.2: + +```sh +PYTHONPATH=installer python3 -m unittest discover -s installer/tests +python3 -m unittest discover -s tools/tests +python3 tools/docs_check.py +python3 tools/manifest_check.py +``` + +Installer: **754 tests, OK, 12 skipped** (81.279 seconds). Tools: **101 tests, OK**. docs_check: **passed**, 218 checked, 8 exempt, 67 warnings. manifest_check: **passed**, policy scan 903 text files, 63 non-UTF-8 skips, zero findings. Additional isolated render inspection confirmed CAS/SP/IdP port and network membership in both modes. Worktree remains unmodified. + +GitHub at review time: `gh pr checks 88` reports **no checks**; check-runs and workflow-run queries return none for this head/branch; combined status is `pending` with an empty statuses array. This is absence of CI evidence, not a passing or actively-running CI result. No prolonged polling was done. Recheck CI and regressions after resolving develop conflicts and R1-01. + +--- + + +## 日本語 + +[→ To English](#english) + +### 判定と範囲 + +**修正要求。** R1-01 はセキュリティ上のマージ阻害事項。R1-02 は所有者判断が必要で、このレビューでは決めない。`gh pr diff 88` と指定 head の新規 detached worktree `rev-pr88` を読んだ。GitHub は `CONFLICTING` であり、develop/S3 の取り込みはコーディネータの担当。マージ・承認・push・commit・grdm-dev スタック操作はしていない。コンテナのビルド・起動や実際の攻撃はしておらず、認証境界の指摘は描画結果と固定済みソースの解析である。実際に偽装ログインしたという報告ではない。 + +### R1-01 — 高: CAS が SP 以外からの未認証の機関認証ヘッダを受け入れる + +箇所: `installer/grdm_dev/templates/cas.json:10`、`installer/grdm_dev/compose.py:683-687`、`installer/grdm_dev/templates/idp/sp-httpd.conf.tmpl:21-22`。固定済みの `RDM-cas-overlay-dev2027`、`22610b7681a29241365108e5129bcd43ade4610e` をローカルソースから変更せずに確認した。対象は `cas-server-support-osf/src/main/java/io/cos/cas/authentication/handler/support/OpenScienceFrameworkPrincipalFromRequestRemoteUserNonInteractiveCredentialsAction.java:347-414`。 + +この action は空でない `AUTH-Shib-Session-ID` を SAML 認証とみなし、要求の `AUTH-*` をコピーして、CAS 自身の署名・暗号化用の秘密を使い、正規化した認証情報を API へ送る。Apache が偽装ヘッダを除去したことを明示的に仮定しており、送信元認証もモードによる無効化もない。`REMOTE_USER` が無くても警告だけで拒否しない。ログインフローは対話式のパスワード認証より前にこの action を呼ぶ。 + +隔離描画では shibboleth の CAS にホストポートは無いが、web・api・rdmcp・IdP ほかと共有する `grdm` に接続したままである。これらの peer は `cas:8080/login` に session・IdP・mail・display name を偽装したヘッダを直接送り、SAML 検証と SP の所属条件を迂回できる。スタイルシートは設定済みのテスト IdP 識別子を受け入れ、独立した所属検証をしない。ホストポートを消すことは必要だが十分ではない。 + +full fakecas では同じ action が SP 無しで `127.0.0.1:8090` に公開される。イメージ既定のスタイルシートは引き続き利用され、`bt` など既知の IdP を対応づけ、`osf_init` はその機関を test セットに登録する。そのためローカルのクライアントは偽装ヘッダで機関認証経路に入れる。その後のアカウント作成・ログインは下流の機関・アカウント条件次第であり、SAML 認証の証明次第ではない。この fakecas の公開経路は本 PR より前からあるが、指定されたレビュー範囲なので報告する。今回の shibboleth 経路でも修復されない。新規 fakecas 描画は新しい mock 用スタイルシートを使わないので、mock IdP の識別子を偽装するだけでは fakecas の有効な再現手順にならない。 + +確認した web/API の CAS クライアントはチケット検証要求と明示した Bearer token の要求を作り、任意の受信ヘッダを転送していない。確認した構成に汎用の web/API → CAS リバースプロキシは見つからなかった。それでも peer からの直接接続は残る。 + +**必要な対策:** CAS が remote credential を作る前に認証境界を強制する。例えば SP → CAS 専用のランダムな共有秘密を生成し、SP がクライアントの同名ヘッダを上書きして backend 要求に付け、CAS が秘密の一致を検査してから認証ヘッダを受け入れる。秘密は実行時に注入し、commit・秘密を含む描画ファイルに入れない。この認証経路は shibboleth のときだけ有効化し、fakecas の全 CAS 入口では機関認証ヘッダを無効化・除去する。認証付きの隔離プロキシ/ネットワーク設計でもよいが、チケット検証用も含めて全 backend 経路を覆う必要がある。SP でのヘッダ消去だけでは SP を通らない要求を守れない。 + +回帰試験には、SP 経由の偽装、通常 peer から CAS への直接要求、full fakecas の 8090、共有秘密なし/不一致、正規の SP 転送を加える。偽装要求で認証済みセッションやチケットが作られないことを検証する。現在の S5 成功/所属条件による拒否では、この境界を試験していない。 + +### R1-02 — 所有者判断: head.js の外部資産例外 + +箇所: `installer/acceptance/netguard.py:46` と SPEC-04・進捗表の対応記述。cdnjs の `/ajax/libs/headjs//head.min.js` を追加したことで、外部要求の失敗ではなく、遮断して数える既知の資産となる。外部通信を許す変更ではないが、受入ゲートを緩める変更である。所有者の明示承認を取得・記録するか、例外を外して head.js をローカル配信する。このレビュワは方針例外を承認も否認もしない。M4 のローカル配信課題は既に記録されている。 + +### その他のセキュリティ・回帰確認 + +- SP は `ShibUseHeaders On` と受信 `REMOTE_USER`/`Remote-User` の早期消去を使い、spoof checking を無効化していない。公式仕様では `checkSpoofing` の既定は true で、SP が管理する受信ヘッダを拒否する。これは SP を通る要求の保護であり、任意の CAS 呼び出し元の保護ではない。任意の名前の `AUTH-*` をすべて消すとの主張でもない。根拠: [Shibboleth SP3 InProcess](https://shibboleth.atlassian.net/wiki/spaces/SP3/pages/2063696243/InProcess)。 +- RequestMap は匿名と member の認証済みセッションを許可し、認証済みの非 member を拒否する。canonical host/port は描画された SP の起点に合う。既存の受入試験は拒否時の SP 403・OSF cookie 不在・API 401、成功時の API の表示名・所属機関を検証する。 +- メタデータには生成した公開証明書を入れ、別々の鍵対を `secrets init` が作る。ディレクトリ 0700、秘密鍵 0600。IdP の salt/admin password は環境変数参照で、起動時と PHP で検査する。固定 admin password はない。mock 利用者のパスワードは意図した fixture である。IdP には自分の秘密鍵だけを mount する。 +- IdP の base は digest 固定、SimpleSAMLphp 2.5.4 tarball は固定 SHA-256 検査付き。描画されたサービスは 1000:1000、read-only、明示 tmpfs、capability drop、no-new-privileges、loopback 限定公開。ビルド・起動を繰り返していないので、構成確認であり、新規の実行時認証ではない。 +- CentOS 7 EOL、amd64 エミュレーション、mod_ssl 省略、M4 の保守された multi-arch base への移行は、英和と進捗表に明記されている。 +- 確認した設計・制約・試験結果は英和で一致し、過去の版の記録も残る。core/minimum は fakecas 選択を維持し、新しい IdP 秘密が必須になるのは full shibboleth のみ。full fakecas は S2 のため CAS を公開したまま。文書化された新規秘密/鍵生成と GRDM_AUTH の保存は S5 以外にも及ぶため、出力が完全に同一との評価ではない。 +- 指定された四つの run フォルダはテキストの証跡のみ。パターン検査と内容確認では実ホスト名・/Users パス・email・token・SAML assertion・秘密鍵は見つからない。JUnit は文字列 `redacted-host`。最新 S5 は 2 passed、最新 full fakecas は 20 passed/4 skipped、先の S5 は 2 passed、先の full は 18 passed/6 skipped。これは提出済みの証跡であり今回のブラウザ再実行ではない。`aipf_mcp_data_grdm` 除外と当時の S3 未実装も記録済み。 + +### 検証 + +`rev-pr88` から Python 3.15.2 で実施した。 + +```sh +PYTHONPATH=installer python3 -m unittest discover -s installer/tests +python3 -m unittest discover -s tools/tests +python3 tools/docs_check.py +python3 tools/manifest_check.py +``` + +Installer は **754 件、OK、12 skipped**(81.279 秒)。Tools は **101 件、OK**。docs_check は **合格**、218 件検査、8 exempt、67 warnings。manifest_check は **合格**、policy scan はテキスト 903 件、非 UTF-8 の除外 63 件、指摘ゼロ。追加の隔離描画で両モードの CAS/SP/IdP のポートとネットワーク所属を確認した。worktree の変更はない。 + +レビュー時の GitHub は `gh pr checks 88` が **no checks**。当該 head/branch の check-runs と workflow runs は空で、combined status は statuses が空の `pending`。CI 合格でも実行中の証拠でもなく、CI 証跡が無い状態である。長時間のポーリングはしていない。develop の競合解消と R1-01 の修正後に CI・回帰を再確認する必要がある。 diff --git a/docs/reviews/2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r2.md b/docs/reviews/2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r2.md new file mode 100644 index 00000000..c65a235b --- /dev/null +++ b/docs/reviews/2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r2.md @@ -0,0 +1,128 @@ +# REV CAS4 r1 and PR88 r2 security review / CAS4 r1・PR88 r2 セキュリティレビュー + +> **Summary:** CAS PR #4 passes the code review. Guide PR #88 must remain on hold until its manifest selects the fixed CAS; its SP boundary and secret wiring are sound. +> **概要:** CAS PR #4 のコードレビューは合格。ガイド PR #88 は修正済み CAS の固定版を manifest に反映するまで保留。SP の認証境界と秘密の配線は妥当。 + +**Jump:** [English](#english) · [日本語](#japanese) + +- Version / 版: 0.1 — 2026-11-16. **Learned:** a tested upstream fix protects an installation only after its pinned dependency selects that fix. / 上流の修正が試験を通っても、固定した依存先がその修正を選ぶまでは配備を保護しない。 +- Applies to: RDM-cas-overlay-dev2027 PR #4 `d9ad5c6661a3796aea63c9da487a9b51bb67b148` (A, r1), and RDM-developer-guide PR #88 `86b7cc752f78326f3edbdfa3b7a2745fe89bb747` (B, r2), both targeting develop. +- Reviewer: REV(担当レビュワ, Codex Astra). +- Related: [Review index](../rev-pr88-r2/docs/reviews/README.md) · [SPEC-06](../rev-pr88-r2/docs/spec/06_documentation_conventions.md) · [SPEC-04](../rev-pr88-r2/docs/spec/04_acceptance_tests.md) · [r1](2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r1.md). Used by: the coordinator's revision and merge decision. + +--- + + +## English + +[→ 日本語版へ](#japanese) + +### Separate verdicts + +**A: PASS for the R1-01 implementation; no blocking code finding.** The only header-derived remote authentication path is guarded before it mutates credentials or calls the institution API. No GitHub approval was submitted. One Japanese r1 comment was posted to CAS #4 (comment 6900937349), and one r2 comment to guide #88 (comment 6900927222). + +**B: HOLD / changes required before merge.** R2-01: `manifest/upstreams.json:460` still selects CAS `22610b7681a29241365108e5129bcd43ade4610e`, which lacks this guard. Merely passing the environment variable to that old implementation does not disable forged authentication in fakecas/core or protect direct network-peer requests. PR #88 explicitly acknowledges this prerequisite. Merge CAS #4 first, update the guide's CAS pin and any canonical mirror, and verify the resulting image/deployment before merging #88. The paired implementation with CAS `d9ad5c66` resolves the reviewed attack; **R1-01 is not yet fully closed for the exact committed guide head**. This is the disclosed integration dependency, not a new defect in the SP wiring. + +Fresh detached clones `rev-cas4` and `rev-pr88-r2` were used. No source changes, commits, pushes, merges, approvals, or grdm-dev stack operations were performed. Only the isolated Maven test container was run. Live login results below are the author's evidence, not a reviewer rerun. + +### A: complete path and header audit + +A whole-overlay search for `REMOTE_USER`, `AUTH-`, `getRemoteUser`, `getHeader`, remote credentials and webflow references found the header consumer in `OpenScienceFrameworkPrincipalFromRequestRemoteUserNonInteractiveCredentialsAction.java`. `login-webflow.xml:98-102` invokes this action from the ordinary login and delegated-login paths. The other `setRemotePrincipal` branches use an existing authenticated PAC4J TGT; the one-time verification path uses its own verification credentials. They do not consume the forged identity headers and should not require the SP secret. + +At lines 368-380 the code checks the trust header before setting SAML delegation, marking a remote principal, reading REMOTE_USER, enumerating identity attributes, invoking the API or issuing a TGT. **Literal-order qualification:** it reads `AUTH-Shib-Session-ID` one line before the trust check and uses its presence to select a generic warning. Thus “no AUTH-* header is read before the check” is not literally true. No value is logged or used to authenticate there; this is a nonblocking wording/order cleanup, not an authentication bypass. + +At lines 888-915, null/blank configuration and an absent header return false; wrong, empty, prefix and different-length values cannot match. The configured value is trimmed. SHA-256 yields two equal-length digests and `MessageDigest.isEqual` compares them without a secret-dependent prefix exit. Hashing the offered value still takes time proportional to its public length; the entire request is not claimed to have constant duration. No trust value or secret is included in the new logs or in the AUTH-* enumeration. + +`getHeader` uses case-insensitive names and returns the first value when a field repeats, per the [Servlet API](https://docs.oracle.com/javaee/7/api/javax/servlet/http/HttpServletRequest.html). Thus a correct first value followed by a wrong one is accepted; a wrong first value followed by the correct one is rejected; a comma-joined value fails equality. This is not a bypass without knowledge of the secret. Rejecting all duplicates and adding explicit case/duplicate tests would make the contract stricter, but are nonblocking. The existing AUTH-* enumeration uses case-sensitive `startsWith`; differently cased attribute names may fail mapping, but cannot bypass the trust check. Apache emits the configured canonical names. + +The added tests exercise absent/wrong headers, unset/null/blank secrets, and the valid control. They assert an error event, no remote flag, no delegation protocol or username, and no `createTicketGrantingTicket` call. Existing SAML success tests now supply the trusted fixture, so fail-closed-only behavior would not satisfy the suite. Explicit case/duplicate integration cases are not included. The default property is empty in `cas-servlet.xml:363`; the existing Spring property-placeholder mechanism supplies environment properties, consistent with the author's live positive control. + +### B: SP order, secrets and evidence + +`secrets init` generates the new key with 32 random bytes encoded by `token_urlsafe` (43 characters), retaining the owner-only `.env` secret store. Reports list key names only. Render passes `${CAS_SP_TRUST_SECRET}` to CAS and SP only in full shibboleth; `$${CAS_SP_TRUST_SECRET}` in the httpd template becomes a literal runtime reference. The startup shell rejects lengths under 32 without printing the value. The intentional secret store `runtime/.env` (and protected rotation backup) is distinct from rendered configuration; runtime container environments necessarily contain the value. + +An independent canary render scanned every output file other than `.env`, plus captured stdout/stderr: no canary value escaped. Shibboleth receivers were exactly `cas` and `shibboleth_sp`; full fakecas and core had none, even when their input `.env` contained the key. Consequently they fail closed **with CAS #4**, not with the current old pin. No committed real secret or secret-bearing run artifact was found. + +`sp-httpd.conf.tmpl:16-44` removes the client trust header, both REMOTE_USER forms, all mapped identity attributes and the CAS session/provider fields in main-server `early` context. It is a list of names, not a wildcard deletion of arbitrary AUTH-* names. Every field consumed by the current stylesheet is covered. Unknown AUTH-* fields cannot create a session, because the session selector is cleared and mod_shib alone restores it for a valid session. Keep the stripping list aligned with future mappings. + +Hook order was checked against [Apache mod_headers semantics](https://httpd.apache.org/docs/2.4/mod/mod_headers.html) and source: early processing occurs before path traversal; late `RequestHeader` runs in fixups before the content handler. `unset` removes duplicate fields, and `set` replaces existing fields case-insensitively. This configuration therefore removes incoming identity values before authentication and sets the server's trust value before proxying, rather than appending to a client value. + +The official [Shibboleth SP 3.4.1 source archive](https://shibboleth.net/downloads/service-provider/3.4.1/shibboleth-sp-3.4.1.tar.gz), `apache/mod_shib.cpp:809-826,978-995,2225-2250`, confirms that `shib_check_user` authenticates and exports headers in the authn hook; `shib_fixups` overlays subprocess environment variables. Apache mod_headers registers early handling at post-read FIRST and late handling at fixups LAST. This establishes the order by hook semantics, not the textual order of unrelated directives. `ShibUseHeaders On` remains, and no `checkSpoofing=false` or spoof-check disabling setting was introduced; its default is true as documented by [Shibboleth InProcess](https://shibboleth.atlassian.net/wiki/spaces/SP3/pages/2063696243/InProcess). These are source/configuration findings, supplemented by the supplied live test, not a new binary-level SP test. + +The author's PR body reports forged GET/POST requests from a web peer, forged SP requests, and full fakecas 8090 requests with both bt and mock identities obtaining no ticket; the configured-secret control obtained a ticket. The committed S5 rejection test uses a fresh context, checks HTTP 200, no ticket redirect, no TGC and API 401 for the forged SP request. The six newly added run folders contain text evidence only; scans found no user-home paths, email addresses, token/ticket assignments, SAML response or private-key material. The latest submitted S5 run has 2 passes; full fakecas has 22 passes and 2 S5 skips. No traces are recorded. The direct-peer/control matrix is PR-body evidence, not a committed automated test matrix. + +### Merge and validation + +Develop `5540c18` is an ancestor. The merge retains #86's OIDC/S3 implementation, #89's aipf pin `51b6f777` and combined-run evidence, and #90's review records. The changed documentation retains every earlier develop version entry, including their text, across both `c77d489` and `5540c18`; the S5 entries are restacked as SPEC-02 0.44, SPEC-04 0.17, status 0.32 and acceptance README 0.9. `git diff --check` passes. R1-02's owner decision is now recorded in both languages; the head.js exception remains pending the separate CAS #5 follow-up. + +From `rev-pr88-r2`: + +```sh +PYTHONPATH=installer python3 -m unittest discover -s installer/tests +python3 -m unittest discover -s tools/tests +python3 tools/docs_check.py +python3 tools/manifest_check.py +``` + +```toon +validation[4]{check,result,detail}: + installer,PASS,"787 tests; 14 skipped; 80.956 s" + tools,PASS,"101 tests" + docs_check,PASS,"223 checked; 8 exempt; 70 warnings" + manifest_check,PASS,"940 text files scanned; 63 non-UTF-8 skipped; zero findings" +``` + +Local Java and Maven are unavailable. The pinned `maven:3.9-eclipse-temurin-8@sha256:0f402712b7a469189cfea5ec69cdcadcb32b22aaf1ea62777ab925c3d8470175` test image ran `mvn -B -s /tmp/review-settings.xml -pl cas-server-support-osf -am test -DskipTests=false` on `git archive HEAD` streamed into its private filesystem. Result: **BUILD SUCCESS**, target action **18 tests**, OSF module **45 tests**, zero failures/errors/skips (44.730 s). The settings file only mirrors repositories to Maven Central; no source was changed. The initial bind mount was not visible in Colima and failed reactor selection; a subsequent archive-based attempt stalled fetching from the old Sonatype endpoint and was stopped before the successful Central-only retry. All reviewer test containers were removed. Evidence: `../rev-cas4-maven-central.log`; guide logs are `../rev-installer.log`, `../rev-tools.log`, `../rev-docs.log`, `../rev-manifest.log`. + +CI snapshot: CAS #4 exposes no checks (`statusCheckRollup=[]`), not a CI pass. Guide #88 has 13 successful checks (Python 3.11/3.12 unit suites, links, MkDocs, shell lint, Ubuntu amd64/arm64 and macOS dry runs, Windows PowerShell, gitleaks, Semgrep, Trivy report-only, manifest policy). The [image recipe job](https://github.com/RCOSDP/RDM-developer-guide/actions/runs/43757904037/job/132945937522) is still in progress. No long CI polling was done. The pin update and remaining CI result remain for the coordinator. + +--- + + +## 日本語 + +[→ To English](#english) + +### 別々の判定 + +**A: R1-01 の実装は PASS。コード上の阻害事項なし。** ヘッダから遠隔認証情報を作る唯一の経路は、credential の変更や機関 API の呼び出し前に保護される。GitHub の approve はしていない。CAS #4 に r1 コメント(6900937349)、ガイド #88 に r2 コメント(6900927222)を日本語で各一件投稿した。 + +**B: HOLD / マージ前の修正が必要。** R2-01: `manifest/upstreams.json:460` はガードの無い CAS `22610b7681a29241365108e5129bcd43ade4610e` を選んだまま。旧実装に環境変数を渡すだけでは、fakecas/core の偽装も peer の直接接続も防げない。PR 本文もこの前提を明記している。CAS #4 を先にマージし、ガイドの固定版と対応する正本を更新し、そのイメージ/配備を確認してから #88 をマージすること。CAS `d9ad5c66` と組み合わせた実装は対象攻撃を解消するが、**ガイドの指定 head の R1-01 はまだ完全には閉じられない**。これは開示済みの依存先統合の残件であり、SP 配線の新しい欠陥ではない。 + +新規 detached clone `rev-cas4`・`rev-pr88-r2` で確認した。ソース変更、commit、push、merge、approve、grdm-dev スタック操作はしていない。隔離した Maven 試験コンテナだけを実行した。実機ログイン結果は作者の証跡であり、レビュワによる再実行ではない。 + +### A: 全経路とヘッダ + +overlay 全体の REMOTE_USER・AUTH-*・getRemoteUser・getHeader・遠隔 credential・webflow の検索で、消費箇所は `OpenScienceFrameworkPrincipalFromRequestRemoteUserNonInteractiveCredentialsAction.java` に集約されていた。`login-webflow.xml:98-102` は通常ログインと委譲ログインから同 action を呼ぶ。他の remotePrincipal 分岐は認証済み PAC4J TGT を使い、一回限りの検証キー経路は独自の資格情報を使う。偽装ヘッダを消費せず、SP 秘密を要求する必要はない。 + +368–380 行の検査は、SAML 設定、remote flag、REMOTE_USER 読み取り、属性列挙、API 呼び出し、TGT 発行より先。**厳密な順序の留保:** AUTH-Shib-Session-ID 自体は検査の一行前に読み、存在を一般的な警告の条件に使う。「検査前には AUTH-* を一切読まない」は文字どおりには正しくない。ただし値のログ出力や認証への利用は無い。順序/表現を整える非阻害の改善点で、認証迂回ではない。 + +888–915 行は秘密が null/空/空白、またはヘッダが欠けると false。誤値、空、接頭辞、長さ違いも一致しない。設定値は trim される。SHA-256 の同じ長さの二つの結果を MessageDigest.isEqual で比較するため、秘密の一致する接頭辞に依存した早期終了が無い。提示値のハッシュ時間は公開された入力の長さに比例し、要求全体の時間が一定という主張ではない。追加ログと AUTH-* 列挙にトラスト値や秘密は入らない。 + +[Servlet API](https://docs.oracle.com/javaee/7/api/javax/servlet/http/HttpServletRequest.html) の getHeader は名前の大文字小文字を区別せず、重複時は最初の値を返す。正値→誤値なら受理、誤値→正値なら拒否、コンマ結合値なら不一致になる。秘密を知らない攻撃者の迂回にはならない。重複を全拒否する実装と大小文字/重複テストは非阻害の改善候補。既存の AUTH-* 列挙は startsWith が大小文字を区別するため、異なる綴りは属性対応に失敗し得るが、信頼検査は迂回できない。Apache は設定どおりの綴りで出力する。 + +追加テストはヘッダ無し/誤値、秘密の未設定/null/空白、正値の対照を検証し、error event、remote flag 無し、protocol/username 無し、TGT 発行呼び出し無しを確認する。既存の SAML 成功試験も正しい秘密付きで維持され、常時拒否するだけでは通らない。大小文字・重複の明示的な統合試験は無い。cas-servlet.xml:363 の既定は空で、既存 Spring property-placeholder による環境変数の解決は作者の正例とも整合する。 + +### B: SP の順序、秘密、証跡 + +secrets init は 32 random bytes を token_urlsafe で 43 文字にする。保存先は所有者専用の .env。報告はキー名だけ。full shibboleth だけが cas と SP に環境変数参照を渡し、httpd テンプレートの `$${CAS_SP_TRUST_SECRET}` は実行時用の `${CAS_SP_TRUST_SECRET}` として残る。起動シェルは 32 文字未満なら値を出さずに拒否する。秘密の保存用 runtime/.env と保護された rotation backup は描画設定とは別であり、実行時のコンテナ環境には当然値が入る。 + +独立した canary 描画で .env 以外の全出力ファイルと stdout/stderr を走査し、値の漏れは無かった。参照先は shibboleth の cas・shibboleth_sp だけ。入力 .env に秘密を残しても full fakecas と core の受取先はゼロ。したがって **CAS #4 なら** fail closed になるが、現在の旧固定版には当てはまらない。commit や run に本物の秘密は見つからなかった。 + +sp-httpd.conf.tmpl:16–44 は main-server の early で、クライアントのトラストヘッダ、二種類の REMOTE_USER、全マッピング属性、CAS の session/provider を消す。これは名前の列挙であり、任意の AUTH-* のワイルドカード消去ではない。現在の stylesheet が使う全項目は覆われる。未知の属性だけでは消去済みの session selector を作れず、有効セッションのときだけ mod_shib が復元する。将来のマッピング追加時にも一覧を同期する必要がある。 + +[Apache 仕様](https://httpd.apache.org/docs/2.4/mod/mod_headers.html) とソースを確認した。early はパスの設定を辿る前、late RequestHeader は content handler の前の fixups。unset は重複を消し、set は大小文字を区別せず既存値を置換する。そのため認証前に client identity を消し、proxy 前にサーバーの秘密を付ける。client の値への append ではない。 + +公式 [SP 3.4.1 ソース](https://shibboleth.net/downloads/service-provider/3.4.1/shibboleth-sp-3.4.1.tar.gz) の apache/mod_shib.cpp:809–826,978–995,2225–2250 では、shib_check_user が authn hook で認証と header export を行い、shib_fixups は subprocess environment を重ねる。mod_headers は post-read FIRST と fixups LAST を使う。単なる設定の行順でなく hook の順序で確認した。ShibUseHeaders On は維持され、spoof checking を無効にする設定は無い。[InProcess の既定](https://shibboleth.atlassian.net/wiki/spaces/SP3/pages/2063696243/InProcess) は true。これはソースと設定の確認を実機証跡で補ったもので、新たな SP バイナリ試験ではない。 + +作者の PR 本文では web peer の偽装 GET/POST、SP 経由の偽装、full fakecas 8090 の bt/mock 身元の偽装はチケット無し、秘密付き対照はチケット有り。commit 済み S5 拒否試験は fresh context で HTTP 200、ticket redirect 無し、TGC 無し、API 401 を確かめる。追加された六つの run はテキストだけで、ホームパス、email、token/ticket 値、SAML response、秘密鍵のパターンは無かった。最新 S5 は 2 pass、full fakecas は 22 pass・S5 の 2 skip。trace は無い。peer/対照の表は PR 本文の実機証跡であり、commit 済み自動試験の一覧ではない。 + +### merge と検証 + +develop 5540c18 は祖先。#86 の OIDC/S3、#89 の aipf 固定版 51b6f777 と統合実行証跡、#90 のレビュー記録が維持される。変更文書の過去の develop 版記録は c77d489・5540c18 の双方から文章ごと残り、S5 は SPEC-02 0.44・SPEC-04 0.17・status 0.32・受入 README 0.9 に積み直されている。git diff --check も合格。R1-02 の所有者判断は英和で記録され、head.js 例外の除去は別件の CAS #5 待ち。 + +rev-pr88-r2 で英語節に記載した四つのコマンドを実行。installer **787 件、14 skipped、80.956 秒、合格**。tools **101 件合格**。docs_check **223 検査、8 exempt、70 warnings、合格**。manifest_check **940 text files、非 UTF-8 除外 63、指摘ゼロ、合格**。 + +ローカル Java/Maven は無い。Dockerfile と同じ digest の Maven 3.9 / Temurin 8 テストイメージに git archive HEAD を渡し、英語節記載の Maven コマンドを実行した。**BUILD SUCCESS、対象 action 18 件、OSF モジュール全体 45 件、失敗・エラー・skip ゼロ、44.730 秒**。設定ファイルで取得先を Maven Central にまとめただけでソース変更は無い。最初の bind mount は Colima から見えず reactor 選択に失敗し、archive を使った試行は旧 Sonatype の取得で停滞したため停止、その後 Central のみで成功した。試験コンテナは全て削除済み。ログは英語節記載の scratchpad ファイルに保存した。 + +CI は CAS #4 の statusCheckRollup が空で、合格とは扱わない。ガイド #88 は Python 3.11/3.12、links、MkDocs、shell lint、Ubuntu 二構成/macOS dry run、PowerShell、gitleaks、Semgrep、Trivy report-only、manifest の **13 checks が成功**。[image recipe job](https://github.com/RCOSDP/RDM-developer-guide/actions/runs/43757904037/job/132945937522) は実行中。長時間の CI ポーリングはしていない。固定版の更新と残りの CI 確認はコーディネータの残件。 diff --git a/docs/reviews/2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r3.md b/docs/reviews/2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r3.md new file mode 100644 index 00000000..97625ac2 --- /dev/null +++ b/docs/reviews/2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r3.md @@ -0,0 +1,66 @@ +# REV PR88 S5 Shibboleth r3 / PR88 S5 Shibboleth レビュー r3 + +> **Summary:** PASS pending pin for the guide delta: the exception removal, run records and bilingual caveats are consistent. Merge still requires the corrected CAS dependency and a pin update. +> **概要:** ガイド差分は PASS pending pin。例外除去・実行記録・英和の制約記載は整合するが、修正した CAS 依存先の統合と固定版更新までマージ不可。 + +**Jump:** [English](#english) · [日本語](#japanese) + +- Version / 版: 0.1 — 2026-11-16. **Learned:** a successful rerun does not erase an initial convergence timeout, and live evidence on an unpinned dependency does not certify the committed pin. / 再実行成功は初回収束タイムアウトを取り消さず、未固定の依存先での実機証拠はコミット済み固定版の証明にならない。 +- Applies to: RDM-developer-guide PR #88 `25cccb29cc5c109afeebe43301d1f30bcaae701b`, delta from r2 `86b7cc752f78326f3edbdfa3b7a2745fe89bb747`. +- Posted comment / 投稿コメント: issuecomment-6901736298. +- Reviewer: REV(担当レビュワ, Codex Astra). +- Related: [Review index](../rev-guide88/docs/reviews/README.md) · [SPEC-06](../rev-guide88/docs/spec/06_documentation_conventions.md) · [r2](2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r2.md) · [CAS r2](2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r2.md). Used by: dependency integration and merge decision. + +--- + + +## English + +[→ 日本語版へ](#japanese) + +### Verdict and prerequisite + +**PASS pending pin for this guide delta, not permission to merge.** No new guide-code blocker was found. The manifest still selects CAS `22610b7681a29241365108e5129bcd43ade4610e`, which lacks both the SP trust fix and the asset cleanup. The live runs explicitly used CAS #5 `ebff4e56ff0489da6aa55d5074721e14f5bf0d1d`, including merged #4 `4cf91e4c`. They do not validate the committed old pin. R2-01's integration hold remains: complete CAS #5, update the manifest and canonical mirrors to its merged commit, and verify that selected image. The paired CAS r2 review requests changes to the new build guard; do not treat this guide verdict as overriding that dependency finding or the documentation's promise that the build rejects any external asset. + +### Delta and evidence + +Only 14 files changed since r2: netguard, its tests, SPEC-04, the status board, and five text files in each of two run folders. No secret-wiring, identity-boundary or acceptance scenario code changed. head.js is removed from KNOWN_EXTERNAL_ASSETS, and the test now rejects its exact former URL as a known exception. Other external assets remain blocked/countable as before. SPEC-04 0.18 and board 0.33 retain earlier versions and describe removal and dependency prerequisites in both languages. + +For acc-20261115T211807Z-6bcf0e, results/collector/JUnit/pytest and regenerated TOON agree: S5 success and rejection pass; HTTP 200/403; known_external_blocked is 7 on success, 0 on rejection. For acc-20261115T212428Z-d13514, all formats agree on 22 passed, 2 skipped (S5 under fakecas), zero failures/errors, and 206 known external blocks. Both use wait_scale 1, no incomplete scenarios and tracing off. The seven success blocks are recorded only as a count, not per-host provenance; attributing them to GRDM pages comes from the author's explanation and source context, not independent per-request evidence. Zero rejection blocks and CAS source inspection support the narrower CAS/SP/IdP observation. + +All ten newly committed files were read/scanned. No personal hostname, /Users or /home path, email address, token/ticket/password/secret assignment, SAML payload or private-key material was found. JUnit retains hostname="redacted-host", a deliberate placeholder. URLs are loopback and pytest documentation. Collector/results test rows and durations agree, JUnit counts agree, and results.toon exactly equals the current encoder output from results.json. No trace or screenshot binary is included. + +### C7 flake judgment + +**A note is necessary and already exists.** The status board's English and Japanese S1–S5 row and the PR body state that the first full run failed C7 at the five-second convergence budget and that a standalone and full rerun passed. Keep that caveat. SPEC-04 §3 rule 2 treats retry-only success as a failed test; the original run remains failed, and the committed green run is subsequent evidence, not proof that C7 became deterministic. This delta does not modify C7 or relax its five-second budget. The disclosed pre-existing flake is a follow-up reliability issue, not a new blocker for removing the CAS exception. Do not report an uninterrupted first-pass suite. The uncommitted failed run was not re-audited here. + +### Checks and scope limits + +From the detached fresh guide clone, `PYTHONPATH=installer python3 -m unittest installer.tests.test_accept` passed 71 tests (10 skipped); docs_check passed 223 files (8 exempt, 71 warnings); manifest_check and `git diff --check 86b7cc7..HEAD` passed. Logs: [unit](../rev-guide88-r3-unit.log), [docs](../rev-guide88-r3-docs.log), [manifest](../rev-guide88-r3-manifest.log). No need to rerun unchanged SP/CAS security wiring; the previous r2 assessment remains applicable, conditioned on the pin. No live stack, secrets, containers of the grdm-dev stack, CI polling, merge, approval, push or commit was performed. CAS's independent build is documented in its separate r2 record. + +--- + + +## 日本語 + +[→ To English](#english) + +### 判定と前提 + +**ガイド差分は PASS pending pin。マージ許可ではない。** 新しいガイド実装の阻害事項はない。manifest は SP 信頼検査と外部資産除去の両方を欠く CAS `22610b7681a29241365108e5129bcd43ade4610e` のまま。実機実行は #4 `4cf91e4c` を含む CAS #5 `ebff4e56ff0489da6aa55d5074721e14f5bf0d1d` を使ったと明示されており、コミット済み旧固定版の検証ではない。R2-01 の統合保留は維持する。CAS #5 を完了し、manifest と対応する正本をそのマージコミットへ更新して選択したイメージを確認すること。同時レビューした CAS r2 は新ビルド検査に修正を要求している。ガイドの判定で依存先の指摘や「任意の外部資産をビルドで拒否する」という文書上の約束を無効にはできない。 + +### 差分と証拠 + +r2 以降の変更は14ファイルだけ。netguard・そのテスト・SPEC-04・進捗ボード、および2実行フォルダー各5個のテキスト。秘密の配線・認証境界・受入シナリオのコード変更はない。KNOWN_EXTERNAL_ASSETS から head.js を除き、旧 URL が例外に該当しないテストへ変更した。他の既知外部資産の遮断・集計は維持される。SPEC-04 0.18 と board 0.33 は旧版を残し、英和両方で除去と依存先の統合前提を記載する。 + +acc-20261115T211807Z-6bcf0e は results・collector・JUnit・pytest・再生成 TOON が整合する。S5 成功・拒否の2件合格、HTTP 200/403、known_external_blocked は成功7・拒否0。acc-20261115T212428Z-d13514 は22件合格・2件 skip(fakecas のため S5)、失敗・エラー0、既知外部遮断206で全形式が整合する。双方 wait_scale 1、未完シナリオなし、trace 無効。成功側7件は件数のみでホスト別の証拠ではない。GRDM 画面由来という帰属は作者の説明とソース上の文脈によるもので、要求単位の独立証拠ではない。拒否側0件と CAS ソース検査は、CAS/SP/IdP に限定した観察を裏付ける。 + +追加された10ファイルを読み、検索した。個人ホスト名、/Users・/home パス、メール、token/ticket/password/secret の値、SAML payload、秘密鍵は見つからなかった。JUnit の hostname="redacted-host" は意図的な置換値。URL は loopback と pytest 文書だけ。collector/results の各テスト行・時間と JUnit 件数は一致し、results.toon は現在の encoder で results.json から生成した内容と完全一致した。trace・画面画像のバイナリは含まれない。 + +### C7 の不安定さの扱い + +**注記は必要であり、既に記録されている。** 進捗ボードの英和 S1–S5 行と PR 本文は、最初の全体実行で C7 が5秒の収束制限を超えて失敗し、単独と全体再実行は成功したことを記載する。この注記を維持すること。SPEC-04 §3 規則2は再試行で通る試験を失敗とみなす。元の実行は失敗のままで、コミットした成功実行は後続の証拠であり、C7 の決定性が回復した証明ではない。今回 C7 のコード・5秒の制限は変更していない。開示済みの既存 flake は信頼性改善の追跡対象だが、CAS 例外除去に新しく生じた阻害事項ではない。初回から中断なしで全体成功したと報告してはならない。未コミットの失敗実行自体は今回再監査していない。 + +### 検査と範囲の限界 + +fresh clone の detached head で `PYTHONPATH=installer python3 -m unittest installer.tests.test_accept` は71件、skip 10で合格。docs_check は223ファイル、除外8、警告71で合格。manifest_check と `git diff --check 86b7cc7..HEAD` も合格。ログは [unit](../rev-guide88-r3-unit.log)・[docs](../rev-guide88-r3-docs.log)・[manifest](../rev-guide88-r3-manifest.log)。変更のない SP/CAS 信頼配線は再試験しておらず、前回 r2 の評価を固定版条件付きで維持する。実機スタック・秘密・grdm-dev コンテナ操作、CI の継続ポーリング、merge・approve・push・commit はしていない。CAS の独立ビルドは別の r2 記録に記載した。 diff --git a/docs/reviews/2026-11-16_RDM-developer-guide_PR89_aipf_Pin_r1.md b/docs/reviews/2026-11-16_RDM-developer-guide_PR89_aipf_Pin_r1.md new file mode 100644 index 00000000..eb77d149 --- /dev/null +++ b/docs/reviews/2026-11-16_RDM-developer-guide_PR89_aipf_Pin_r1.md @@ -0,0 +1,92 @@ +# REV PR89 aipf pin review r1 / aipf 固定版レビュー r1 + +> **Summary:** Independent review of the aipf pin and the combined acceptance evidence; no blocking defect found, with runtime image provenance limited to the author's report. +> **概要:** aipf 固定版と同時受入の証跡を独立に確認した。阻害する問題は見つからず、実行イメージの由来は著者の報告に限られる。 + +**Jump:** [English](#english) · [日本語](#japanese) + +- Version / 版: 0.1 — 2026-11-16. **Learned:** acceptance results establish scenario outcomes but do not by themselves record the running image identities. / 受入結果はシナリオの成否を示すが、それだけでは実行したイメージの識別子を記録しない。 +- Applies to / 対象: RCOSDP/RDM-developer-guide PR #89, head `9c299c8422e0a537908741bc1550d17feab74400`, base `c77d48977e3b1c0329f31a7631eb964fd5d18085`. +- Reviewer / 担当: REV(担当レビュワ, Codex Astra). +- Related / 関連: [Review index](../rev-pr89/docs/reviews/README.md) · [SPEC-06](../rev-pr89/docs/spec/06_documentation_conventions.md) · [Status board](../rev-pr89/docs/status/README.md). +- Used by / 利用先: [PR #89 review decision](https://github.com/RCOSDP/RDM-developer-guide/pull/89). + +--- + + +## English + +[→ 日本語版へ](#japanese) + +### Verdict + +**PASS with a non-blocking evidence limitation.** No revision is required for the pin change. This is a review verdict, not GitHub approval or authorization to merge; unfinished CI still needs its normal completion. + +### Evidence + +Read `gh pr diff 89` and inspected a fresh detached `rev-pr89` worktree at the exact head. The change is limited to one manifest SHA, status-board updates and five acceptance artifacts. + +The GitHub API reports the aipf repository's develop tip as `51b6f77722196d927954f3e77b0454beb4c442f8`, also the merge commit of its PR #1 (merged 2026-11-15T18:45:32Z). The comparison from `94e277da` changes only `src/aipf_mcp_data_grdm/server.py` and `tests/test_access_log.py`. + +Both board languages report run `acc-20261115T190938Z-0d3f6d`, 22 passed, zero failed and two S5 skips. These match results.json, collector.json, junit.xml and pytest.log; results.toon is exactly reproduced by the report encoder from results.json. Earlier versions 0.30 and below and earlier scenario histories remain. The full-run precheck records 27 healthy services; the board's 23 services refers to the later restored core profile, not that precheck. + +**R1-01 / P3, non-blocking:** the four board image tags agree with manifest pins and the oidc-mock recipe tag (`rdmcp:75151e455db6`, `aipf_mcp_data_grdm:51b6f7772219`, `oidc-mock:2`, `oasys:b3860fda5172`). However, none of the five run files records image identities or the post-run core health report. Actual loaded images, no-service-stop history and post-run health exit 0 remain author-reported. Consider attaching a sanitized image/health inventory for stronger future reproducibility; this review does not claim that the run folder independently proves those facts. + +All five run files were read and searched for user paths, email addresses, JWT shapes and credential fields. No actual machine hostname, `/Users` path, email, JWT, token value or secret was found. JUnit retains only `hostname="redacted-host"`; `invalid_token` and `wrong_audience` are error labels, and loopback addresses are not private host identities. No binary attachments or traces are present. + +At base develop `c77d489`, Shibboleth-SP has no build_platform. The builder selects the host platform absent an override, includes it in full builds, and maps build failure to exit 30. Its pinned Dockerfile uses CentOS 7 and explicit x86_64 RPMs. This supports the PR's arm64 limitation and the need for linux/amd64; the failed real build itself was not rerun or independently witnessed. + +### Validation and remaining CI + +Commands were run from `rev-pr89`; expected result is exit 0: + +```sh +python3 tools/manifest_check.py +python3 tools/docs_check.py +PYTHONPATH=installer python3 -m grdm_dev.cli manifest toon --check +PYTHONPATH=installer python3 -m unittest discover -s installer/tests +``` + +Manifest validation passed. Documentation validation passed for 219 files with 8 exemptions and 70 warnings. Manifest TOON check passed. Installer suite passed: 773 tests, 14 skipped, 98.866 seconds on Python 3.15.2. The initial invocation using `-m grdm_dev` was invalid because the package has no `__main__`; the corrected CLI invocation above passed. + +CI snapshot: 10 checks passed (MkDocs, links, manifest/private-URL policy, gitleaks, Trivy, PowerShell, shellcheck, macOS dry run and both Ubuntu dry runs). Four were pending: image builds, Semgrep and installer/tools unit jobs for Python 3.11 and 3.12. No failed check was present; no long polling was performed. Acceptance was reviewed from supplied artifacts, not rerun against the live stack. No source edits, commits, pushes, approval or merge were performed. + +--- + + +## 日本語 + +[→ To English](#english) + +### 判定 + +**PASS(非阻害の証跡上の制限あり)。** 固定版変更について必須修正は無い。これはレビュー判定であり GitHub 承認やマージ許可ではない。未完了の CI は通常どおり完了確認が必要。 + +### 確認した証拠 + +`gh pr diff 89` と、対象先頭の新しい detached worktree `rev-pr89` を確認した。変更は manifest の SHA 一つ、進捗ボード、受入証跡5ファイルに限られる。 + +GitHub API で aipf リポジトリの develop 先頭は `51b6f77722196d927954f3e77b0454beb4c442f8`。PR #1 のマージコミットと一致し、マージ日時は 2026-11-15T18:45:32Z。`94e277da` との差分は `src/aipf_mcp_data_grdm/server.py` と `tests/test_access_log.py` だけ。 + +ボード両言語の実行 ID `acc-20261115T190938Z-0d3f6d`、合格22、失敗0、S5 のスキップ2は results.json、collector.json、junit.xml、pytest.log と一致する。results.json から既存のレポート生成器で results.toon を完全に再現できた。0.30 以下の版と過去のシナリオ履歴は保持されている。full 実行前の27サービス正常と、ボードの core 復帰後23サービスは異なる時点を指す。 + +**R1-01 / P3、非阻害:** ボードのイメージ4件は manifest の固定版と oidc-mock のレシピ版に一致する(`rdmcp:75151e455db6`、`aipf_mcp_data_grdm:51b6f7772219`、`oidc-mock:2`、`oasys:b3860fda5172`)。ただし実行フォルダ5ファイルにイメージ識別子や core 復帰後の健康確認記録は無い。実際の使用イメージ、サービスを止めなかった履歴、実行後の health 終了0は著者報告として扱う。今後の再現性を強めるには、秘匿化したイメージ・健康確認一覧の添付が望ましい。実行フォルダだけでこれらを独立に証明したとは扱わない。 + +実行フォルダ全5ファイルを読み、利用者パス、メール、JWT 形式、認証情報の項目を検索した。実機ホスト名、`/Users` パス、メール、JWT、トークン値、秘密値は見つからない。JUnit は `hostname="redacted-host"` のみを残し、`invalid_token` と `wrong_audience` はエラー名。ループバックアドレスは実機固有のホスト情報ではない。バイナリ添付やトレースは無い。 + +基点 develop `c77d489` の Shibboleth-SP に build_platform は無い。builder は指定が無いとホストのプラットフォームを選び、full の対象に含め、ビルド失敗を終了30にする。固定版 Dockerfile は CentOS 7 と明示的な x86_64 RPM を使う。PR の arm64 制限と linux/amd64 指定の必要性を裏付けるが、実ビルド失敗そのものを再実行・目撃したわけではない。 + +### 検証と残る CI + +実行場所は `rev-pr89`、期待値は終了0。 + +```sh +python3 tools/manifest_check.py +python3 tools/docs_check.py +PYTHONPATH=installer python3 -m grdm_dev.cli manifest toon --check +PYTHONPATH=installer python3 -m unittest discover -s installer/tests +``` + +manifest 検証は合格。文書検証は219ファイル、例外8、警告70で合格。manifest TOON 検査も合格。installer suite は Python 3.15.2 で773件、スキップ14、98.866秒で合格。最初の `-m grdm_dev` は `__main__` が無いため実行できなかったが、正しい CLI 呼出しで合格した。 + +CI 確認時点は10件合格(MkDocs、リンク、manifest/private-URL、gitleaks、Trivy、PowerShell、shellcheck、macOS dry run、Ubuntu の両 dry run)。image builds、Semgrep、Python 3.11 と3.12の installer/tools unit jobs の4件は未完了。失敗は無く、長時間のポーリングはしていない。受入は提出証跡の確認であり実スタックでの再実行ではない。ソース編集、コミット、push、承認、マージは行っていない。 diff --git a/docs/reviews/2026-11-16_RDM-developer-guide_PR91_osf_Pin_c5a05de_r1.md b/docs/reviews/2026-11-16_RDM-developer-guide_PR91_osf_Pin_c5a05de_r1.md new file mode 100644 index 00000000..aeb9ea6a --- /dev/null +++ b/docs/reviews/2026-11-16_RDM-developer-guide_PR91_osf_Pin_c5a05de_r1.md @@ -0,0 +1,101 @@ +# REV PR91 osf.io pin review r1 / osf.io 固定版レビュー r1 + +> **Summary:** No blocking findings at PR #91 head 368793d; the pin is suitable for integration once the remaining image-build check succeeds. +> **概要:** PR #91 の先頭 368793d に阻害指摘なし。残るイメージのビルド検査が成功すれば統合に適した固定版と判断する。 + +**Jump:** [English](#english) · [日本語](#japanese) + +- Version / 版: 0.1 — 2026-11-16. **Learned:** the acceptance run proves core compatibility, while READ-user authorization needs separate upstream code and test evidence. / core の受入は互換性を示すが、READ の権限判定には上流のコードと試験という別の根拠が要る。 +- Applies to: RCOSDP/RDM-developer-guide PR #91, head `368793d0b4677ad2c31eac4f342c69611b03c59c`, base `79ba329f538377378cd9f855856f97fbb70db605`. +- Reviewer: REV(担当レビュワ, Codex Astra); independent round 1. +- Related: [review index](../rev-pr91/docs/reviews/README.md) · [SPEC-06](../rev-pr91/docs/spec/06_documentation_conventions.md) · [PR #91](https://github.com/RCOSDP/RDM-developer-guide/pull/91). +- Used by: [DEMO-20](../rev-pr91/docs/demo/20_core_install_mac_studio.md) · [status board](../rev-pr91/docs/status/README.md). + +--- + + +## English + +[→ 日本語版へ](#japanese) + +### Verdict and scope + +PASS with no blocking findings. Pinning is safe within the reviewed scope; the outstanding image-recipes CI check remains a merge gate. No merge, approval, commit, push or stack operation was performed. Review used `gh pr diff 91` and a fresh detached worktree `scratchpad/rev-pr91` at the exact head. + +### Upstream and security + +The GitHub branch API returned `c5a05de0e3d007558566e3262b26b2da38e76d49` as RDM-osf.io-dev2027 develop tip. The comparison from `07cc56d47b3922e12dedc7fc55b86b1014ffc5e6` is 5 ahead, 0 behind, with seven changed files and no migration. The commits are `041de092` (authorization), `6a14ce29` (race correction), `e16aa6b2` (tests), `236fc621` (merge #39), and `c5a05de0` (merge #40). PR #40 changes only `api_tests/nodes/views/test_private_mapcore_children.py`. Push Unit Tests run `43754596744` and CodeQL run `43754596779` both succeeded; every Unit Tests job, including private_security, website and addons, succeeded. + +I read the complete #39 diff and the pinned handlers, helper, permission decorators and TimestampTask model. `add_timestamp_token` requires WRITE before queueing or replacing a task. `cancel_task` requires WRITE or the checked task's requester, rejects anonymous users, and passes that same row to the cancellation helper. The helper revokes that task ID and deletes only the matching primary key plus task ID. Verification without WRITE locks the existing row, refuses another user's running task, and handles an absent-row race using the node's OneToOne constraint and IntegrityError before dispatching work. Status and page cleanup also delete only the task they inspected. IQB-RIMS checks WRITE immediately before its `?edit=` write; an existing edit state or absent edit argument remains a read-only page path. WRITE addition, cancellation, verification and IQB-RIMS editing retain their successful paths and have upstream regression tests. + +No remaining unauthorized READ-user timestamp-add, other-requester cancellation/replacement, or IQB-RIMS edit-state write path was found. This is deliberately narrower than “READ users cause no database writes”: verification, own-request cancellation, completed-task cleanup and download audit logging remain permitted by the upstream decision. The upstream tests exercise refusals, successful WRITE operations, requester exceptions and injected races, but are not a real multi-connection/Celery concurrency test. I did not rerun the upstream application suite locally; the pinned push CI and static inspection provide that evidence. + +### Evidence and documentation + +The five files in `acc-20261116T003758Z-d077f9` agree: results JSON re-encodes exactly to TOON, collector rows agree with results, and JUnit/log counts are 16 passed, 8 skipped, no failures/errors. Core C1–C7 and S1 pass; S2–S5 skip for the stated profile/auth reasons. HTTP rejection statuses, 189 blocked known external requests, 23 prechecked services, wait scale 1.0 and 108751 ms overall duration match DEMO-20. The 85.260-second JUnit/pytest duration is the test subprocess duration, not a contradiction of the 109-second overall run. + +DEMO-20 0.26 has its new English section before the Japanese anchor and Japanese section before the glossary. Board 0.37 agrees in both languages; prior version entries in both files are preserved. The docs explicitly say the READ-user refusals were not exercised in this acceptance run and #41 is excluded. Build durations, image recreation and restoration are operator narrative; this acceptance-only folder does not independently attest those operations or contain a running-image SHA inventory. + +All five run files were inspected and scanned for local paths, hostnames, emails and credential markers. No actual hostname, `/Users/` path, email, token or secret was found. JUnit retains only the literal placeholder `hostname="redacted-host"`; the base URL is loopback and evidence paths are repository-relative. PR gitleaks also passed. + +### Validation + +Commands ran from the fresh worktree using local Python 3.15: + +```sh +PYTHONPATH=installer python3 -m unittest discover -s installer/tests +python3 -m unittest discover -s tools/tests +python3 tools/manifest_check.py +python3 tools/docs_check.py +PYTHONPATH=installer python3 -m grdm_dev.cli manifest toon --check +rg -n '07cc56d4' installer/tests tools/tests +``` + +```toon +checks[6]{check,result}: + installer,"787 tests, OK, 14 skipped, 80.036 s" + tools,"101 tests, OK" + manifest_check,"PASS; manifest and tracked-file policy have zero findings" + docs_check,"PASS; 223 files, 8 exemptions, 71 non-failing warnings" + manifest_toon,"up to date" + old_pin_search,"no matches in installer/tests or tools/tests" +``` + +The initially attempted `python3 -m grdm_dev` is not a supported entry point; the successful check above uses `grdm_dev.cli`. Installer tests emitted a Python 3.15 fork deprecation and a foreign local listener warning, without failures. + +Latest PR CI snapshot: 13 checks passed (Python 3.11/3.12 tests, MkDocs, links, manifest/policy, gitleaks, Semgrep, Trivy, shell lint and four OS dry-run checks); only “Build image_build recipes (amd64 + arm64, no push)” remained pending at run `43782073606`. No long CI polling was performed. Remaining action: coordinator confirms that image check succeeds before normal merge. + +--- + + +## 日本語 + +[→ To English](#english) + +### 判定と範囲 + +阻害指摘なしの PASS。確認した範囲で固定版の採用は安全と判断するが、残る image-recipes の CI 成功はマージ条件として残る。マージ・承認・コミット・push・スタック操作は行っていない。`gh pr diff 91` と、正確な先頭を checkout した新規の detached worktree `scratchpad/rev-pr91` で独立確認した。 + +### 上流とセキュリティ + +GitHub のブランチ API は RDM-osf.io-dev2027 develop の先頭として `c5a05de0e3d007558566e3262b26b2da38e76d49` を返した。旧固定版 `07cc56d47b3922e12dedc7fc55b86b1014ffc5e6` から 5 コミット先、遅れ 0、変更 7 ファイルで migration は無い。内訳は `041de092`(権限判定)、`6a14ce29`(競合修正)、`e16aa6b2`(試験)、`236fc621`(#39 マージ)、`c5a05de0`(#40 マージ)。#40 は `api_tests/nodes/views/test_private_mapcore_children.py` だけを変える。push の Unit Tests 実行 `43754596744` と CodeQL 実行 `43754596779` は成功し、Unit Tests は private_security・website・addons を含む全ジョブが成功した。 + +#39 の全差分、固定版のハンドラ・補助関数・権限デコレータ・TimestampTask モデルを読んだ。`add_timestamp_token` はキュー投入や行の置換より前に WRITE を要求する。`cancel_task` は WRITE または確認したタスクの依頼者を要求し、匿名を拒否し、同じ行を取消の補助関数へ渡す。取消はその task ID に限定され、行の削除も主キーと task ID の一致が条件となる。WRITE の無い利用者による検証は既存行をロックし、ほかの人の実行中タスクを拒否する。行が無い場合の競合は node の OneToOne 制約と IntegrityError で、処理を投入する前に拒否する。状態取得・ページの片付けも、見たタスクだけを削除する。IQB-RIMS は `?edit=` の書き込み直前に WRITE を検査し、edit が既にある場合や引数が無い場合は読み取りだけとなる。WRITE の付与・取消・検証・IQB-RIMS 編集の成功経路は残り、上流に回帰試験がある。 + +READ による無権限のタイムスタンプ付与、他人のタスク取消・置換、IQB-RIMS の編集状態更新は見つからなかった。ただし「READ はデータベースに一切書かない」という意味ではない。検証、自分の依頼の取消、完了済みタスクの片付け、ダウンロードの監査ログは上流の決定により許されている。上流試験は拒否、WRITE の成功、依頼者の例外、割り込ませた競合を確認するが、実際の複数接続・Celery 並行実行試験ではない。上流のアプリケーション試験は手元で再実行せず、固定 SHA の push CI と静的確認を根拠とした。 + +### 証拠と文書 + +`acc-20261116T003758Z-d077f9` の 5 ファイルは整合する。results JSON から再生成した TOON は完全一致し、collector の各行も results と一致する。JUnit とログも合格 16、スキップ 8、失敗・エラー 0。core の C1〜C7 と S1 が合格し、S2〜S5 は記載した profile/auth の理由で飛ばされる。拒否状態、既知の外部要求の遮断 189、事前確認 23 サービス、wait scale 1.0、全体時間 108751 ms は DEMO-20 と一致する。JUnit/pytest の 85.260 秒は試験プロセスの時間で、全体 109 秒との矛盾ではない。 + +DEMO-20 0.26 の英語の新節は日本語アンカーより前、日本語の新節は用語集より前にある。ボード 0.37 は両言語で一致し、両ファイルの過去の版は保存されている。READ の拒否そのものは今回の受入で試していないこと、#41 は含まないことも明記されている。ビルド時間・イメージの作り直し・復元は作業者の記録であり、受入だけのフォルダからは独立に証明できず、実行イメージの SHA 一覧も入っていない。 + +5 ファイルを読み、ローカルパス・ホスト名・メール・資格情報の目印も検索した。実ホスト名、`/Users/` パス、メール、token、secret は見つからなかった。JUnit の hostname 属性は `redacted-host` という置換文字列だけ。base URL は loopback、証拠パスはリポジトリ相対である。PR の gitleaks も成功した。 + +### 検証 + +英語節のコマンドを新規 worktree でローカル Python 3.15 により実行した。installer は 787 件、80.036 秒、スキップ 14 で OK。tools は 101 件で OK。manifest_check は manifest と追跡ファイルの方針検査の指摘 0 で PASS。docs_check は 223 ファイル、例外 8、失敗にしない警告 71 で PASS。`manifest toon --check` は最新と判定した。installer/tests と tools/tests に旧 SHA の一致は無い。 + +最初に試した `python3 -m grdm_dev` は対応する入口ではなく、成功した検査は `grdm_dev.cli` を使っている。installer 試験では Python 3.15 の fork 非推奨警告と無関係なローカル listener の警告が出たが、失敗は無い。 + +最後の PR CI 確認では 13 検査が成功した(Python 3.11/3.12、MkDocs、リンク、manifest/方針、gitleaks、Semgrep、Trivy、shell lint、4 種の OS dry-run)。実行 `43782073606` の “Build image_build recipes (amd64 + arm64, no push)” だけが継続中。長い CI 待機はしていない。残りは coordinator による同検査の成功確認と、その後の通常マージ判断である。 diff --git a/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r1.md b/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r1.md new file mode 100644 index 00000000..bd936a20 --- /dev/null +++ b/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r1.md @@ -0,0 +1,71 @@ +# aipf PR1 Access-log review r1 / アクセスログのレビュー r1 + +> **Summary:** CHANGES REQUESTED: an encoded query parameter name bypasses redaction. Ordinary token logging is fixed and the submitted tests pass. +> **概要:** エンコードされたパラメータ名でマスクを回避できるため CHANGES REQUESTED。通常のトークンは隠れ、提出された試験は合格する。 + +**Jump:** [English](#english) · [日本語](#japanese) + +- Version / 版: 0.1 — 2026-11-16. **Learned:** log redaction must recognize the same parameter names as the request parser. / ログのマスクは要求の解析器と同じパラメータ名を認識しなければならない。 +- Applies to: RCOSDP/RDM-aipf_mcp_data_grdm-dev2027 PR #1, `0f472a024192459adf68454730cfae0e07e3d8e8`, base develop. +- Related: [Review index](../rev-pr86-r2/docs/reviews/README.md) · [SPEC-06](../rev-pr86-r2/docs/spec/06_documentation_conventions.md) · [Source](../rev-aipf-pr1/src/aipf_mcp_data_grdm/server.py) · [Tests](../rev-aipf-pr1/tests/test_access_log.py). +- Used by: PR #1 revision and coordinator review. + +--- + + +## English + +[→ 日本語版へ](#japanese) + +### Findings + +**CHANGES REQUESTED. R1-01 (medium, required): encoded parameter names leak the token.** `server.py:334–339` matches only literal `t=`. `/grdm/token?%74=VALUE` is parsed by Starlette as parameter `t`, but `redact_query` leaves VALUE intact. Independently reproduced with a real loopback uvicorn h11 server using `log_config()`: the handler received the decoded `t` value and the access log contained the synthetic secret. This is a valid alternative spelling of the same credential-bearing request, not merely a malformed log line. Match decoded query parameter names and redact every matching value while preserving unrelated parameters; add both function-level and live access-log regression cases for encoded names and repeated encoded/plain names. A valid form token can be replayed for its remaining 600-second lifetime to replace the subject's stored PAT, as identified in guide PR #86 r1. + +**R1-02 (low, compatibility follow-up): combined request-line records are not sanitized.** The five-argument `(client, method, path, version, status)` record is handled. A three-argument `(client, request_line, status)` record using `%s - "%s" %d`, or a preformatted message, remains unchanged and can retain the credential. Neither shape caused a logging exception in the checks. Current installed uvicorn 0.54.0 h11 and httptools implementations both emit the supported five-argument shape, so this is not evidence of another default-runtime leak. Either document this supported shape or extend handling and tests if alternate access-log record formats are part of the intended contract. + +### Checks + +Repeated literal `t` parameters are all masked. Other parameters before/after `t` survive. Percent-encoded values are masked; the failing case is the encoded name. `at=` is untouched. Routes without `t` stay unchanged; `/other?t=...` is also masked, which is a conservative extension beyond `/grdm/token`, not a correctness blocker. Empty, mapping and single-argument unrelated log records still format successfully. The filter returns true and preserves status/other arguments. `log_config` deep-copies uvicorn defaults, adds the filter to the access handler and is passed by `main`; the default configuration is not mutated. + +Fresh clone and isolated venv under the scratchpad. `uv` and Python 3.12 were unavailable, so the AGENTS.md install/test workflow was performed using `python3 -m venv` and pip, on Python 3.15.2 (within the declared >=3.10 support range). No hooks were installed and no source changes were made. + +```toon +checks[5]{check,result}: + pytest,"9 passed in 2.91s; fresh editable installation with test extras" + secret_scan,"python3 tools/scan_secrets.py --all: passed" + supplemental_matrix,"5 query cases and 3 unrelated record shapes passed" + live_encoded_name,"decoded t received=true; synthetic secret leaked=true" + github_checks,"no checks reported for PR branch" +``` + +The tests cover one live uvicorn backend and an ordinary five-argument LogRecord; they do not cover encoded parameter names or the combined request-line shape. Fix R1-01 and add those security regression cases before integration. No stack operation, approval, merge, commit or push occurred. + +--- + + +## 日本語 + +[→ To English](#english) + +### 指摘 + +**CHANGES REQUESTED。R1-01(中・必須):エンコードされたパラメータ名からトークンが漏れる。** `server.py:334–339` は文字どおりの `t=` だけを認識する。`/grdm/token?%74=VALUE` を Starlette は `t` と解釈するが、`redact_query` は VALUE を残す。`log_config()` を使う本物の loopback uvicorn h11 で、ハンドラがデコード後の `t` を受け取り、アクセスログに合成の秘密文字列が残ることを独立に再現した。これは同じ資格情報を送る有効な別表記であり、壊れたログだけの問題ではない。query の名前をデコードしたうえで全ての該当値を隠し、無関係なパラメータは保持する必要がある。エンコードされた名前、通常名との重複を、関数試験と実ログ試験に追加してほしい。有効なフォームトークンは残りの有効時間(最大 600 秒)中に再利用して対象 subject の PAT を置き換えられる(guide PR #86 r1)。 + +**R1-02(低・互換性の追跡):要求行をまとめたレコードは隠れない。** 五引数 `(client, method, path, version, status)` は処理されるが、`%s - "%s" %d` の三引数 `(client, request_line, status)` や整形済みメッセージは無変更で、資格情報が残り得る。今回の確認ではログの例外は起きない。導入された uvicorn 0.54.0 の h11 と httptools はどちらも対応済みの五引数を使うため、既定の実行時にさらに漏れる証拠ではない。対応する形式を明記するか、別形式も契約に含むなら処理と試験を追加してほしい。 + +### 検査 + +通常の `t` が二度現れても両方隠れる。前後の別パラメータは保持する。値の percent encoding は隠れるが、名前の encoding が今回の不具合。`at=` は変えない。`t` のない経路は変えず、`/other?t=...` も隠すのは保守的な範囲拡大であり阻害事項ではない。引数なし、mapping、一引数の無関係なレコードも正常に整形できる。フィルタは true を返し、status と他の引数を保持する。`log_config` は既定設定を deep copy して access handler にフィルタを加え、`main` から渡すため、既定設定自体は書き換えない。 + +scratchpad の新規 clone と独立 venv を使用した。`uv` と Python 3.12 が無いため、AGENTS.md の導入・試験を `python3 -m venv` と pip で実行した。Python は 3.15.2(宣言された >=3.10 の範囲内)。hooks は導入せず、ソース変更も行っていない。 + +```toon +checks[5]{check,result}: + pytest,"新規 test extras 導入; 9 passed; 2.91 秒" + secret_scan,"python3 tools/scan_secrets.py --all: 合格" + supplemental_matrix,"query 5 ケース、無関係なレコード 3 形式は合格" + live_encoded_name,"デコードされた t を受信=true; 合成の秘密値の漏洩=true" + github_checks,"PR ブランチにチェック報告なし" +``` + +既存試験は一つの実 uvicorn backend と通常の五引数 LogRecord を扱うが、エンコード名と要求行をまとめた形式を扱わない。統合前に R1-01 とセキュリティ回帰試験の対応が必要。スタック操作、approve、merge、commit、push はしていない。 diff --git a/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r2.md b/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r2.md new file mode 100644 index 00000000..b27eb104 --- /dev/null +++ b/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r2.md @@ -0,0 +1,73 @@ +# aipf PR1 Access-log review r2 / アクセスログのレビュー r2 + +> **Summary:** CHANGES REQUESTED for a new formatting-error leak; both r1 findings are fixed. The 23 submitted tests pass and detect the old implementation. +> **概要:** r1 の両指摘は解決したが、新しい整形エラーによる漏洩のため CHANGES REQUESTED。提出された 23 試験は合格し、旧実装を検出する。 + +**Jump:** [English](#english) · [日本語](#japanese) + +- Version / 版: 0.1 — 2026-11-16. **Learned:** rewriting a logging format string must preserve its argument contract or the logging error path can disclose the original arguments. / ログの書式文字列を書き換えるとき引数との対応を壊すと、ログのエラー処理から元の引数が漏れる。 +- Applies to: RCOSDP/RDM-aipf_mcp_data_grdm-dev2027 PR #1, head `01820f0`, base develop; REV round 2. +- Related: [Review index](../rev-pr86-r2/docs/reviews/README.md) · [SPEC-06](../rev-pr86-r2/docs/spec/06_documentation_conventions.md) · [r1](2026-11-16_aipf_PR1_AccessLog_r1.md) · [Source](../rev-aipf-pr1/src/aipf_mcp_data_grdm/server.py). +- Used by: PR #1 revision and coordinator decision. + +--- + + +## English + +[→ 日本語版へ](#japanese) + +### Verdict and finding + +**CHANGES REQUESTED. R2-01 (medium): do not remove interpolation placeholders while retaining positional arguments.** In `RedactFormToken.filter`, the `record.msg = redact_text(record.msg)` assignment can erase `%s` from a valid parameterized log message. Reproduction: a LogRecord equivalent to `logger.info('POST /grdm/token?t=%s', synthetic_token)`. The filter produces message `POST /grdm/token?[redacted]` but retains `(synthetic_token,)`. `getMessage()` raises TypeError, and a standard StreamHandler's error path writes the original `Arguments: (...)` to stderr when `logging.raiseExceptions` has its default true value. Independently captured that stderr and confirmed the synthetic token is present. This both loses the intended access-log line and discloses the value through diagnostic logging. + +This is an additional ordinary parameterized logging shape, not the five-argument record emitted by current uvicorn, which passes. Preserve formatting semantics while sanitizing the final sensitive content, and add a real handler/stderr regression for a URL embedded in the message template with positional substitution. Do not rely on globally disabling logging exceptions: that hides the leak by dropping the line rather than fixing the filter. Keep compatibility with uvicorn's AccessFormatter, which uses its structured five arguments. + +### Prior findings and bypass checks + +R1-01 is resolved for the actual form endpoint: its entire query is removed, including `%74`, `%54`, repeated names, mixed order and encoded values. R1-02 is resolved for the submitted five-argument, combined three-argument request line, dictionary and preformatted message cases. Tests and prior reproduction shapes now redact successfully. + +Additional parser-aware checks: + +- `/grdm/token?%2574=...`: entire query is removed; Starlette decodes the name once to `%74`, not `t`. +- `/grdm/token?x=1;t=...`: entire query is removed; current Starlette does not parse the semicolon as a separate `t` parameter. +- `/grdm/token.?%74=...`: route does not match `/grdm/token`; the fallback still masks the decoded `t` value. +- `/grdm/%2574oken?%2574=...`: one decode leaves `/grdm/%74oken` and parameter `%74`, so the route and credential parameter do not match. The text is left intact, but this is not a bypass to the current endpoint. No double-decoding proxy is part of the reviewed direct uvicorn deployment. +- Unrelated startup, five-argument access, three-argument request-line, mapping and preformatted logs without sensitive parameters retain exactly the same rendered text (five independent cases). Other query parameters remain on unrelated routes; `t`/`T` masking there is intentional. Query removal on the form route is intentionally broader. + +### Validation and limits + +Existing isolated clone and venv from r1, detached at the requested head, Python 3.15.2; no stack operations. Full pytest: 23 passed (2.35 seconds initially, 2.05 seconds after restoration). Secret scan `python3 tools/scan_secrets.py --all` passed. Temporarily replaced only server.py with r1 `0f472a0`, leaving current tests: 16 failed and seven passed, including failure of the real uvicorn test. Restored HEAD, reran all tests successfully, and confirmed clean git status/diff. The added handler probe reproduces R2-01 outside the committed suite. + +GitHub reports no checks on the branch. No commit, push, approval or merge occurred. One review comment was posted. Required remaining work: fix R2-01 and add its regression, then rerun tests. + +--- + + +## 日本語 + +[→ To English](#english) + +### 判定と指摘 + +**CHANGES REQUESTED。R2-01(中):位置引数を残したまま書式の置換記号を消さない。** `RedactFormToken.filter` の `record.msg = redact_text(record.msg)` は、正しいログの書式から `%s` を消す。`logger.info('POST /grdm/token?t=%s', synthetic_token)` 相当の LogRecord で再現した。フィルタ後のメッセージは `POST /grdm/token?[redacted]` だが、引数 `(synthetic_token,)` は残る。`getMessage()` は TypeError になり、標準 StreamHandler のエラー処理は `logging.raiseExceptions` が既定の true なら元の `Arguments: (...)` を stderr へ出す。stderr を独立に捕捉し、合成のトークンが含まれることを確認した。意図したログ行が失われ、診断ログから値が漏れる。 + +これは通常の引数付きログの追加の形であり、今回合格した現行 uvicorn の五引数レコードとは異なる。書式と引数の対応を守って秘密情報を隠し、URL を書式文字列に含め位置引数で値を渡すケースを、実 handler と stderr を使う回帰試験に追加してほしい。ログ例外を全体で無効化して済ませると、行を捨てて問題を隠すだけになる。構造化した五引数を使う uvicorn AccessFormatter との互換性も維持する必要がある。 + +### 前回指摘と回避の検査 + +R1-01 は実際のフォーム endpoint について解決した。query 全体を消すので、`%74`、`%54`、重複、混在した順番、値の encoding も隠れる。R1-02 も提出された五引数、要求行をまとめた三引数、辞書、整形済みメッセージについて解決した。前回の再現形式も現在は隠れる。 + +解析器との対応を確認した追加ケース: + +- `/grdm/token?%2574=...`:query 全体が消える。Starlette の一度のデコードでは名前は `%74` となり、`t` ではない。 +- `/grdm/token?x=1;t=...`:query 全体が消える。現行 Starlette はセミコロンを別の `t` パラメータの区切りとして扱わない。 +- `/grdm/token.?%74=...`:`/grdm/token` の route には一致しないが、fallback でデコード後の `t` の値は隠れる。 +- `/grdm/%2574oken?%2574=...`:一度のデコード後は `/grdm/%74oken` とパラメータ `%74` のため、経路と資格情報名の両方が不一致。文字列は残るが、現在の endpoint への回避ではない。今回の直接 uvicorn 構成に二重デコードする proxy は含まれない。 +- 無関係な起動メッセージ、五引数アクセス、三引数要求行、mapping、整形済みログの五ケースは出力文字列が完全に一致した。他経路では無関係な query を保持し、`t`/`T` のマスクは意図したもの。フォーム経路で query 全体を消すのも意図した範囲である。 + +### 検証と制限 + +r1 の独立 clone/venv を指定 head の detached checkout として使用。Python 3.15.2、スタック操作なし。pytest 全体は 23 passed(初回 2.35 秒、復元後 2.05 秒)。`python3 tools/scan_secrets.py --all` は合格。現行の試験を残して server.py だけを r1 `0f472a0` に一時的に戻すと 16 failed、7 passed で、実 uvicorn 試験も失敗した。HEAD を復元して全試験を再実行し合格、git status/diff が clean であることを確認した。追加の handler 検査はコミット済み試験の外で R2-01 を再現している。 + +GitHub のブランチにはチェック報告なし。commit、push、approve、merge はしていない。レビューコメントは一件投稿。残りは R2-01 の修正、回帰試験の追加、試験再実行。 diff --git a/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r3.md b/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r3.md new file mode 100644 index 00000000..2b5e1207 --- /dev/null +++ b/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r3.md @@ -0,0 +1,85 @@ +# aipf PR1 Access-log review r3 / アクセスログのレビュー r3 + +> **Summary:** PASS: R2-01 is resolved, regression tests detect the previous implementation, and normal uvicorn formatting is preserved. +> **概要:** R2-01 は解決し、回帰試験は旧実装を検出し、通常の uvicorn 整形も維持されるため PASS。 + +**Jump:** [English](#english) · [日本語](#japanese) + +- Version / 版: 0.1 — 2026-11-16. **Learned:** a sanitized rendered message with empty arguments preserves literal percent signs without a second formatting pass. / 秘匿済みの整形結果と空の引数なら、再整形せずに percent 記号をそのまま保てる。 +- Applies to: RCOSDP/RDM-aipf_mcp_data_grdm-dev2027 PR #1, head `25bf430`, base develop; REV round 3. +- Related: [Review index](../rev-pr86-r2/docs/reviews/README.md) · [SPEC-06](../rev-pr86-r2/docs/spec/06_documentation_conventions.md) · [r2](2026-11-16_aipf_PR1_AccessLog_r2.md) · [Source](../rev-aipf-pr1/src/aipf_mcp_data_grdm/server.py). +- Used by: PR #1 integration decision. + +--- + + +## English + +[→ 日本語版へ](#japanese) + +### Verdict + +**PASS. R2-01 resolved; no new blocking finding.** Re-ran the original real StreamHandler reproduction of `logger.info('POST /grdm/token?t=%s', synthetic_token)`: the output is redacted, stderr is empty and the token is absent. The mapping variant also passes. Added independent handler checks for split template/argument URLs, an invalid argument count/type, a literal percent in another query parameter, and mapping substitution. All five cases emitted no token or stderr diagnostic. + +The structured fast path only retains arguments after rendering the cleaned arguments succeeds and the resulting text needs no further redaction. Otherwise the original record is rendered, the resulting text is redacted, and args is cleared. Failed rendering produces the marker without appending argument values. This closes the observed placeholder/argument mismatch and the resulting logging-error disclosure. The review addresses credentials carried in the form URL; it is not a claim that arbitrary unrelated exception payloads or custom formatters are universal secret-safe channels. + +### Percent signs and AccessFormatter + +Inspected the running Python `logging.LogRecord.getMessage`: it calls `str(self.msg)`, then performs `msg % self.args` only under `if self.args`. The empty tuple is false, so args=() correctly avoids a second formatting pass. Independently checked a rendered `/other?x=100%&t=[redacted]`: the percent remains single and no formatting error occurs. Escaping it to `%%` would incorrectly change this already-rendered message. + +Current uvicorn's native five-argument records stay structured. Compared filtered versus unfiltered AccessFormatter output for `/healthz`, `/mcp?x=1&at=2`, and `/other?percent=100%25`: all three lines are byte-identical and retain five arguments. The submitted real AccessFormatter test also verifies the expected redacted form-route line with client, method, HTTP version and status intact. Generic/preformatted and malformed records use the fallback as designed; it is not a claim that uvicorn's specialized five-argument formatter accepts arbitrary record shapes it never supported. + +Reran prior parser-aware checks: the form route's whole query is removed for a double-encoded name or semicolon-separated text; a trailing-dot route is not the form route but its decoded t value is still masked. A double-encoded path/name does not reach the form endpoint under the application's single decoding and is not an authenticated-route bypass. No remaining disclosure was found in the reviewed supported URL/logging paths. + +### Validation + +Existing isolated scratchpad clone/venv, detached at the requested head, Python 3.15.2. No application stack operation. + +```toon +checks[6]{check,result}: + full_pytest,"36 passed; 2.20s initially, 1.89s after restoration" + mutation_previous_01820f0,"7 failed, 29 passed; current tests detect the pre-fix implementation" + independent_handlers,"5 cases passed; no token in output or stderr" + normal_AccessFormatter,"3 lines byte-identical; five arguments retained" + secret_scan,"python3 tools/scan_secrets.py --all: passed" + github_checks,"no checks reported on the branch" +``` + +For mutation verification only server.py was temporarily replaced with `01820f0`; restored HEAD, reran the full suite, and confirmed clean git status and diff. No commit, push, approve or merge. One review comment posted; no code correction remains from this review. + +--- + + +## 日本語 + +[→ To English](#english) + +### 判定 + +**PASS。R2-01 は解決し、新しい阻害指摘はない。** 前回の実 StreamHandler による `logger.info('POST /grdm/token?t=%s', synthetic_token)` を再実行し、出力は秘匿され、stderr は空で、トークンが無いことを確認した。mapping 版も合格。独立に、URL が書式と引数にまたがる場合、不正な引数個数・型、別 query の literal percent、mapping 置換を検査し、五ケースすべてでトークンも stderr 診断も出なかった。 + +構造を保持する経路は、引数の秘匿後に整形でき、結果をさらに秘匿する必要がない場合だけ引数を残す。それ以外は元のレコードを整形し、結果を秘匿して args を空にする。整形不能なら引数値を並べずにマーカーを出す。これにより、確認済みの置換記号と引数の不整合、およびログエラーからの漏洩は解決する。対象はフォーム URL に含まれる資格情報であり、無関係な例外 payload や任意の custom formatter まで万能に秘密を除去するという判断ではない。 + +### Percent 記号と AccessFormatter + +実行中の Python の `logging.LogRecord.getMessage` を確認した。`str(self.msg)` の後、`if self.args` が真の場合だけ `msg % self.args` を行う。空 tuple は偽なので、args=() は再整形を正しく避ける。`/other?x=100%&t=[redacted]` を独立に確認し、percent が一つのまま残り、整形エラーも起きないことを確かめた。ここで `%%` にすると、整形済みのメッセージを誤って変えてしまう。 + +現行 uvicorn の通常の五引数レコードは構造を保持する。`/healthz`、`/mcp?x=1&at=2`、`/other?percent=100%25` についてフィルタ有無の AccessFormatter 出力を比較し、三行とも byte 単位で一致し、引数も五つ残る。提出された実 AccessFormatter 試験は、秘匿後のフォーム行についても client、method、HTTP version、status を含む期待出力を確認する。一般・整形済み・不正レコードは設計どおり fallback を使う。uvicorn 専用の五引数 formatter が、もともと非対応の任意形式を受理するとの判断ではない。 + +前回の解析器との比較も再実行した。二重 encoding の名前やセミコロンを含むフォーム query は全体が消える。末尾ドットはフォーム経路とは不一致だが、デコードした t の値は隠れる。二重 encoding のパス・名前はアプリの一度の decoding ではフォームへ到達せず、認証経路の回避ではない。確認対象の URL・ログ経路で残る漏洩は見つからなかった。 + +### 検証 + +既存の独立 scratchpad clone/venv を指定 head の detached checkout で使用。Python 3.15.2。アプリのスタック操作なし。 + +```toon +checks[6]{check,result}: + full_pytest,"36 passed; 初回 2.20 秒、復元後 1.89 秒" + mutation_previous_01820f0,"7 failed、29 passed; 修正前を現行試験で検出" + independent_handlers,"5 ケース合格; 出力・stderr にトークンなし" + normal_AccessFormatter,"3 行 byte 単位で一致; 五引数を保持" + secret_scan,"python3 tools/scan_secrets.py --all: 合格" + github_checks,"ブランチにチェック報告なし" +``` + +mutation 確認では server.py だけを一時的に `01820f0` に戻した。HEAD 復元後に全試験を再実行し、git status/diff が clean であることを確認した。commit、push、approve、merge は行っていない。コメントは一件投稿し、このレビューからのコード修正残件はない。 From 8d4ff02063911f0e601852049f77fdb6008c3c98 Mon Sep 17 00:00:00 2001 From: Yusuke-KOMIYAMA Date: Mon, 16 Nov 2026 13:13:48 +0900 Subject: [PATCH 2/3] =?UTF-8?q?docs(review):=20=E5=8F=8E=E3=82=81=E3=81=9F?= =?UTF-8?q?=E3=83=AC=E3=83=93=E3=83=A5=E3=83=BC=E8=A8=98=E9=8C=B2=E3=81=AE?= =?UTF-8?q?=E3=83=AA=E3=83=B3=E3=82=AF=E3=81=A8=E5=9B=BA=E6=9C=89=E3=81=AE?= =?UTF-8?q?=E3=83=91=E3=82=B9=E3=82=92=E7=9B=B4=E3=81=99=EF=BC=88PR=20#92?= =?UTF-8?q?=20=E3=83=AC=E3=83=93=E3=83=A5=E3=83=BC=20r1=20R1-01=E3=80=9CR1?= =?UTF-8?q?-03=EF=BC=89?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit - R1-01: docs_check が見つけたリンク切れ 186 件を直した。リポジトリの文書を指すもの 160 件は docs/reviews/ からの正しい相対パスにし(チェックアウトの接頭辞を外したときにできた ..docs/ の 壊れた形を含む)、scratchpad のログや diff など、リポジトリに無い証跡 26 件は、リンクを外して 「external evidence not kept in the repository」と明示した。証跡を作り足してはいない。 - R1-02: PR22 r1 の非公開リポジトリの URL を、リポジトリ名と PR・run の番号の書き方にした(SPEC-01 規則 5)。 - R1-03: 一時のチェックアウトとログのパス /private/tmp/... を /... に、VM のホームのパスを / にした。 判定と指摘の中身は変えていない。docs_check、manifest_check、installer と tools の unittest は合格。 Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BW1YQk71jLXY2FaSzMnnYJ --- ...RDM-developer-guide_PR57_Records_Live_ZAP_r1.md | 8 ++++---- ...loper-guide_PR60_Wait_Scale_Trivy_Timeout_r1.md | 14 +++++++------- ...loper-guide_PR60_Wait_Scale_Trivy_Timeout_r2.md | 10 +++++----- ...eveloper-guide_PR61_M2_Close_ADR_Approval_r1.md | 8 ++++---- ...eveloper-guide_PR61_M2_Close_ADR_Approval_r2.md | 4 ++-- ...RDM-developer-guide_PR62_Amd64_Final_Pins_r1.md | 4 ++-- ...RDM-developer-guide_PR62_Amd64_Final_Pins_r2.md | 8 ++++---- ...-developer-guide_PR63_Scrub_Host_Identity_r1.md | 4 ++-- ...-developer-guide_PR63_Scrub_Host_Identity_r2.md | 8 ++++---- ...developer-guide_PR64_Install_Volume_Guard_r1.md | 4 ++-- ...DM-developer-guide_PR65_Workflow_Scope_CI_r1.md | 4 ++-- ...DM-developer-guide_PR65_Workflow_Scope_CI_r2.md | 4 ++-- ...M-developer-guide_PR68_Status_Publication_r1.md | 2 +- ...M-developer-guide_PR69_DEMO20_macOS_Fresh_r1.md | 2 +- ...4_RDM-ember-osf-web-dev2027_PR5_Node24_CI_r1.md | 2 +- ...M-osf.io-dev2027_PR22_Private_Security_CI_r1.md | 6 +++--- ..._RDM-developer-guide_PR75_osf_Pin_94e6817_r1.md | 2 +- ..._RDM-developer-guide_PR76_Up_Stale_Render_r1.md | 4 ++-- ...5_RDM-developer-guide_PR78_S1_y_websocket_r1.md | 2 +- ...5_RDM-developer-guide_PR78_S1_y_websocket_r2.md | 2 +- ...-developer-guide_PR82_DEMO22_Empty_Docker_r1.md | 2 +- ...6-11-15_RDM-developer-guide_PR83_S4_OASys_r1.md | 2 +- ...6-11-15_RDM-developer-guide_PR83_S4_OASys_r2.md | 2 +- ..._RDM-developer-guide_PR84_osf_Pin_07cc56d_r1.md | 2 +- ...-16_RDM-cas-overlay_PR5_NoExternalScripts_r1.md | 6 +++--- ...-16_RDM-cas-overlay_PR5_NoExternalScripts_r2.md | 6 +++--- ...-16_RDM-cas-overlay_PR5_NoExternalScripts_r3.md | 6 +++--- ...-16_RDM-cas-overlay_PR5_NoExternalScripts_r4.md | 10 +++++----- ...16_RDM-developer-guide_PR85_Health_Settle_r1.md | 2 +- ...16_RDM-developer-guide_PR85_Health_Settle_r2.md | 2 +- ...-16_RDM-developer-guide_PR86_S3_aipf_OIDC_r1.md | 2 +- ...-16_RDM-developer-guide_PR86_S3_aipf_OIDC_r2.md | 2 +- ...6-11-16_RDM-developer-guide_PR87_S2_rdmcp_r1.md | 6 +++--- ...16_RDM-developer-guide_PR88_S5_Shibboleth_r1.md | 2 +- ...16_RDM-developer-guide_PR88_S5_Shibboleth_r2.md | 2 +- ...16_RDM-developer-guide_PR88_S5_Shibboleth_r3.md | 6 +++--- ...6-11-16_RDM-developer-guide_PR89_aipf_Pin_r1.md | 2 +- ..._RDM-developer-guide_PR91_osf_Pin_c5a05de_r1.md | 4 ++-- docs/reviews/2026-11-16_aipf_PR1_AccessLog_r1.md | 2 +- docs/reviews/2026-11-16_aipf_PR1_AccessLog_r2.md | 2 +- docs/reviews/2026-11-16_aipf_PR1_AccessLog_r3.md | 2 +- 41 files changed, 87 insertions(+), 87 deletions(-) diff --git a/docs/reviews/2026-11-13_RDM-developer-guide_PR57_Records_Live_ZAP_r1.md b/docs/reviews/2026-11-13_RDM-developer-guide_PR57_Records_Live_ZAP_r1.md index e964bad0..53d7496d 100644 --- a/docs/reviews/2026-11-13_RDM-developer-guide_PR57_Records_Live_ZAP_r1.md +++ b/docs/reviews/2026-11-13_RDM-developer-guide_PR57_Records_Live_ZAP_r1.md @@ -28,7 +28,7 @@ ### Record comparison and live-run evidence -All seven originals were read from the supplied review directory and diffed against the added publication files. [Full diff](PR57_r1_records.diff) contains only the two added version/provenance lines, repository link destinations, and unavailable local-log/audit links converted to plain references. A normalization comparison passed for PR50 r1/r2, PR52 r1/r2 and PR56 r1/r2/r3; no judgment, finding or test result was rewritten. Their original 0.1 lines remain. This proves fidelity, not sanitization: R1-01 remains. An initial audit assertion was corrected because it mistakenly treated linked repository JSON as a local log; the corrected comparison passed all seven. +All seven originals were read from the supplied review directory and diffed against the added publication files. Full diff (`PR57_r1_records.diff`, external evidence not kept in the repository) contains only the two added version/provenance lines, repository link destinations, and unavailable local-log/audit links converted to plain references. A normalization comparison passed for PR50 r1/r2, PR52 r1/r2 and PR56 r1/r2/r3; no judgment, finding or test result was rewritten. Their original 0.1 lines remain. This proves fidelity, not sanitization: R1-01 remains. An initial audit assertion was corrected because it mistakenly treated linked repository JSON as a local log; the corrected comparison passed all seven. DEMO-20 adds 0.22, keeps all 22 prior Learned lines (0.21 only loses its repeated Version prefix), and retains English before Japanese with corresponding additions. Commit `ca2a0105877393e4c297a0175fa3139f2ab7c7bd` is the PR56 merge. The committed scanner implements Python inside the ZAP image, one GET without following redirects, matching the configured sign-in destination, a skip note, and exclusion from scanned-target count. The M1 manifest and gate code support report-only semantics. These code facts do not prove that the live invocation succeeded. @@ -47,7 +47,7 @@ checks[5]{command,exit,result}: "git diff --check origin/develop...HEAD",0,"no whitespace errors" ``` -Logs: [docs](PR57_r1_docs.log), [manifest](PR57_r1_manifest.log), [installer](PR57_r1_installer.log), [tools](PR57_r1_tools.log). Installer output includes expected negative-case messages, a fork deprecation warning and a foreign-listener warning; the final result and exit code pass. Skips and documentation warnings remain limitations. Final Git status is clean. No live runtime/.env access, Docker state changes, merge, approval, close, push or commit occurred; tests use temporary fixtures and the Compose integration check is read-only config. One concise Japanese PR comment accompanies this report. Author corrections and re-review remain; no live scan rerun is required merely to qualify the evidence honestly. +Logs: docs (`PR57_r1_docs.log`, external evidence not kept in the repository), manifest (`PR57_r1_manifest.log`, external evidence not kept in the repository), installer (`PR57_r1_installer.log`, external evidence not kept in the repository), tools (`PR57_r1_tools.log`, external evidence not kept in the repository). Installer output includes expected negative-case messages, a fork deprecation warning and a foreign-listener warning; the final result and exit code pass. Skips and documentation warnings remain limitations. Final Git status is clean. No live runtime/.env access, Docker state changes, merge, approval, close, push or commit occurred; tests use temporary fixtures and the Compose integration check is read-only config. One concise Japanese PR comment accompanies this report. Author corrections and re-review remain; no live scan rerun is required merely to qualify the evidence honestly. --- @@ -66,7 +66,7 @@ Logs: [docs](PR57_r1_docs.log), [manifest](PR57_r1_manifest.log), [installer](PR ### 原本比較とライブ実行の証拠 -指定レビュー directory の原本7件と公開版を diff した。[全差分](PR57_r1_records.diff) は版・出典の2行追加、リポジトリ内リンク先変更、未公開 log・audit リンクの平文参照化だけだった。PR50 r1/r2、PR52 r1/r2、PR56 r1/r2/r3 の正規化比較は全件一致し、判断・指摘・試験結果の書換えはない。元の0.1行も保持される。ただし原文忠実性と秘匿化は別であり R1-01 は残る。最初の監査 assertion はリポジトリの JSON リンクをローカル log と誤分類したため修正し、正しい分類では7件とも一致した。 +指定レビュー directory の原本7件と公開版を diff した。全差分 (`PR57_r1_records.diff`, external evidence not kept in the repository) は版・出典の2行追加、リポジトリ内リンク先変更、未公開 log・audit リンクの平文参照化だけだった。PR50 r1/r2、PR52 r1/r2、PR56 r1/r2/r3 の正規化比較は全件一致し、判断・指摘・試験結果の書換えはない。元の0.1行も保持される。ただし原文忠実性と秘匿化は別であり R1-01 は残る。最初の監査 assertion はリポジトリの JSON リンクをローカル log と誤分類したため修正し、正しい分類では7件とも一致した。 DEMO-20 は0.22を追加し、過去の Learned 22行すべてを保持する(0.21 は重複する Version 接頭辞だけ除去)。英語→日本語と追加内容の対応も維持する。`ca2a0105877393e4c297a0175fa3139f2ab7c7bd` は PR56 のマージである。コードには ZAP イメージ内の Python、一回の GET、転送を追わない動作、設定済みサインイン宛先との照合、skip 注記、検査済み件数からの除外が実装されている。M1 manifest とゲート処理は report-only の意味を裏付けるが、実スタックでの成功はコードだけでは証明できない。 @@ -85,4 +85,4 @@ checks[5]{command,exit,result}: "git diff --check origin/develop...HEAD",0,"空白エラーなし" ``` -ログ: [docs](PR57_r1_docs.log)、[manifest](PR57_r1_manifest.log)、[installer](PR57_r1_installer.log)、[tools](PR57_r1_tools.log)。installer には想定内の否定系メッセージ、fork 非推奨警告、foreign-listener 警告があるが、最終結果と終了コードは成功。skip と文書警告は限界として残る。最終 Git status は clean。稼働環境の runtime/.env 閲覧、Docker 状態変更、merge・approve・close・push・commit は行っていない。試験は一時 fixture を使い Compose 統合検査は読取り専用 config である。日本語の PR コメントを1件添付する。著者の修正と再レビューが残るが、証拠の位置づけを正すためだけにライブ検査再実行は要求しない。 +ログ: docs (`PR57_r1_docs.log`, external evidence not kept in the repository)、manifest (`PR57_r1_manifest.log`, external evidence not kept in the repository)、installer (`PR57_r1_installer.log`, external evidence not kept in the repository)、tools (`PR57_r1_tools.log`, external evidence not kept in the repository)。installer には想定内の否定系メッセージ、fork 非推奨警告、foreign-listener 警告があるが、最終結果と終了コードは成功。skip と文書警告は限界として残る。最終 Git status は clean。稼働環境の runtime/.env 閲覧、Docker 状態変更、merge・approve・close・push・commit は行っていない。試験は一時 fixture を使い Compose 統合検査は読取り専用 config である。日本語の PR コメントを1件添付する。著者の修正と再レビューが残るが、証拠の位置づけを正すためだけにライブ検査再実行は要求しない。 diff --git a/docs/reviews/2026-11-13_RDM-developer-guide_PR60_Wait_Scale_Trivy_Timeout_r1.md b/docs/reviews/2026-11-13_RDM-developer-guide_PR60_Wait_Scale_Trivy_Timeout_r1.md index 235d7305..86171ae6 100644 --- a/docs/reviews/2026-11-13_RDM-developer-guide_PR60_Wait_Scale_Trivy_Timeout_r1.md +++ b/docs/reviews/2026-11-13_RDM-developer-guide_PR60_Wait_Scale_Trivy_Timeout_r1.md @@ -7,7 +7,7 @@ - Version / 版: 0.1 — 2026-11-13. **Learned:** scaling inner waits does not extend an outer process deadline, and inherited settings can make recorded evidence disagree with execution. / 内側の待ちを伸ばしても親プロセスの期限は伸びず、継承した設定で実行と証跡が食い違うと分かった。 - Applies to: RCOSDP/RDM-developer-guide PR #60, head `29cb4ffe1bc32125d39e990137cec2377afb4bb9`, base `develop`; REV round 1. -- Related: [Review index](../../../../../pr60-rev-20261113/docs/reviews/README.md) · [SPEC-06](../../../../../pr60-rev-20261113/docs/spec/06_documentation_conventions.md) · [SPEC-04](../../../../../pr60-rev-20261113/docs/spec/04_acceptance_tests.md) · [PR #60](https://github.com/RCOSDP/RDM-developer-guide/pull/60) +- Related: [Review index](README.md) · [SPEC-06](../spec/06_documentation_conventions.md) · [SPEC-04](../spec/04_acceptance_tests.md) · [PR #60](https://github.com/RCOSDP/RDM-developer-guide/pull/60) - Used by: PR #60 revision and coordinator review decision. --- @@ -19,7 +19,7 @@ ### Verdict and findings -**Changes requested.** Review completed against the pinned head in the clean clone `/private/tmp/pr60-rev-20261113`. No repository files were modified, committed or pushed. No Docker configuration or runtime/.env was changed. This report is outside the repository. +**Changes requested.** Review completed against the pinned head in the clean clone `/pr60-rev-20261113`. No repository files were modified, committed or pushed. No Docker configuration or runtime/.env was changed. This report is outside the repository. 1. **R1-01 — P2: the outer pytest deadline is not scaled.** `installer/grdm_dev/accept.py:217–220,716`, related to `installer/acceptance/conftest.py:61–62`. The default runner still passes `timeout=1900` to `subprocess.run`, and the pytest invocation supplies no override. At scale 20 the inner suite permits 36,000 seconds and even a single test permits 2,400 seconds, yet the parent kills the process after 1,900 seconds. A fake subprocess audit through the real default runner reproduced `[1900]` with `wait_scale=20.0`; no actual long wait was needed. Derive the pytest process deadline from the effective scale, retain cleanup allowance and unchanged scale-1 behavior, and test the real runner boundary. 2. **R1-02 — P2: an inherited scale silently changes execution while results claim 1.** `installer/grdm_dev/accept.py:687–691,747`. `Context.env` copies the host environment; copying it again and omitting a new key does not remove an existing `GRDM_ACCEPT_WAIT_SCALE`. With that key set to `20` and no CLI flag, the existing fake-runner fixture produced exit 0, child environment `20`, and `results.json.wait_scale=1.0`. The suite reads the inherited value and raises waits and limits. Remove the inherited key before applying an explicitly supplied flag, and use a single effective value for execution and evidence. Cover polluted parent environments and explicit override. The current test always starts with `ctx.env={}`, so it misses this case. @@ -39,7 +39,7 @@ The committed run `acc-20261113T024626Z-b2178f` has `wait_scale=3.0`, seeded exe ### Validation -Commands ran from `/private/tmp/pr60-rev-20261113`, using host `python3` (Homebrew Python 3.15.2, outside any virtual environment). All completed: +Commands ran from `/pr60-rev-20261113`, using host `python3` (Homebrew Python 3.15.2, outside any virtual environment). All completed: ```sh python3 tools/docs_check.py @@ -58,7 +58,7 @@ checks[4]{command,exit,result}: tools_tests,0,"93 tests; 3.543 seconds" ``` -Logs: `/private/tmp/pr60-check-0.log` through `/private/tmp/pr60-check-3.log`, in command order. Additional focused reproductions and the fixture audit exited 0. Tests use fakes; no live acceptance or scanner run was performed. The remaining work is to address R1-01 through R1-05 and rerun focused regressions plus required checks. +Logs: `/pr60-check-0.log` through `/pr60-check-3.log`, in command order. Additional focused reproductions and the fixture audit exited 0. Tests use fakes; no live acceptance or scanner run was performed. The remaining work is to address R1-01 through R1-05 and rerun focused regressions plus required checks. --- @@ -69,7 +69,7 @@ Logs: `/private/tmp/pr60-check-0.log` through `/private/tmp/pr60-check-3.log`, i ### 判定と指摘 -**修正要求(Changes requested)。** クリーンクローン `/private/tmp/pr60-rev-20261113` の指定 head をレビューした。リポジトリ内のファイル変更・コミット・push は行っていない。Docker 設定と runtime/.env は変更していない。レビュー記録はリポジトリ外に置いた。 +**修正要求(Changes requested)。** クリーンクローン `/pr60-rev-20261113` の指定 head をレビューした。リポジトリ内のファイル変更・コミット・push は行っていない。Docker 設定と runtime/.env は変更していない。レビュー記録はリポジトリ外に置いた。 1. **R1-01 — P2: pytest の親プロセスの期限が倍にならない。** `installer/grdm_dev/accept.py:217–220,716` と `installer/acceptance/conftest.py:61–62`。既定 runner は `subprocess.run` に `timeout=1900` を渡し続け、pytest 呼び出しも上書きしない。倍率 20 では内部の全体制限が 36,000 秒、1 件が 2,400 秒でも、親が 1,900 秒で終了させる。実際の既定 runner を通す偽 subprocess で、倍率 20 でも `[1900]` を再現した。長時間の実待機は不要だった。pytest の親期限を有効倍率から計算し、後処理の余裕と倍率 1 の既定動作を維持すること。実 runner 境界の試験を追加すること。 2. **R1-02 — P2: 継承した倍率が未指定でも効き、結果は 1 と記録される。** `installer/grdm_dev/accept.py:687–691,747`。`Context.env` はホスト環境を複製するため、新しいキーを追加しないだけでは既存の `GRDM_ACCEPT_WAIT_SCALE` は消えない。そのキーを `20` にしてフラグを省略すると、既存の偽 runner で終了 0・子環境 `20`・`results.json.wait_scale=1.0` を再現した。スイートは継承値で待ちと制限を伸ばす。継承キーを除去してから明示フラグを設定し、実行と証跡に同じ有効値を使うこと。汚染した親環境と明示上書きの試験を追加すること。現在の試験は `ctx.env={}` なので検出できない。 @@ -89,7 +89,7 @@ CLI の倍率解析は 1〜20 を受け付け NaN・無限大を拒否する。 ### 検証 -`/private/tmp/pr60-rev-20261113` から、仮想環境外のホスト `python3`(Homebrew Python 3.15.2)で次を実行した。 +`/pr60-rev-20261113` から、仮想環境外のホスト `python3`(Homebrew Python 3.15.2)で次を実行した。 ```sh python3 tools/docs_check.py @@ -108,4 +108,4 @@ checks[4]{command,exit,result}: tools_tests,0,"93 件; 3.543 秒" ``` -ログは実行順に `/private/tmp/pr60-check-0.log` から `/private/tmp/pr60-check-3.log`。追加の再現と fixture 検査も終了 0。試験は偽物を使い、実スタックの受入・走査は実行していない。残作業は R1-01〜R1-05 の修正と、回帰試験・必須検査の再実行である。 +ログは実行順に `/pr60-check-0.log` から `/pr60-check-3.log`。追加の再現と fixture 検査も終了 0。試験は偽物を使い、実スタックの受入・走査は実行していない。残作業は R1-01〜R1-05 の修正と、回帰試験・必須検査の再実行である。 diff --git a/docs/reviews/2026-11-13_RDM-developer-guide_PR60_Wait_Scale_Trivy_Timeout_r2.md b/docs/reviews/2026-11-13_RDM-developer-guide_PR60_Wait_Scale_Trivy_Timeout_r2.md index 19f10746..16494d90 100644 --- a/docs/reviews/2026-11-13_RDM-developer-guide_PR60_Wait_Scale_Trivy_Timeout_r2.md +++ b/docs/reviews/2026-11-13_RDM-developer-guide_PR60_Wait_Scale_Trivy_Timeout_r2.md @@ -7,7 +7,7 @@ - Version / 版: 0.1 — 2026-11-13. **Learned:** checking the child environment together with the real parent runner and saved results establishes that the selected scale reaches all three boundaries. / 子環境・実際の親 runner・保存結果を一緒に確認することで、指定倍率が三つの境界に届くと確かめられる。 - Applies to: RCOSDP/RDM-developer-guide PR #60, base `develop`, head `f3964dd2efd389fbe5dd3107600f3e05c83f7894`; REV(担当レビュワ), round 2. -- Related: [SPEC-06](../../../../../pr60-rev-r2-f3964dd/docs/spec/06_documentation_conventions.md) · [Round 1](2026-11-13_RDM-developer-guide_PR60_Wait_Scale_Trivy_Timeout_r1.md) · [PR #60](https://github.com/RCOSDP/RDM-developer-guide/pull/60) +- Related: [SPEC-06](../spec/06_documentation_conventions.md) · [Round 1](2026-11-13_RDM-developer-guide_PR60_Wait_Scale_Trivy_Timeout_r1.md) · [PR #60](https://github.com/RCOSDP/RDM-developer-guide/pull/60) --- @@ -30,7 +30,7 @@ ### Validation and limits -Host `/opt/homebrew/bin/python3` is Python 3.15.2, outside a virtual environment. Docs and unit checks ran from the pinned archive `/private/tmp/pr60-rev-r2-f3964dd`. The full manifest check needs Git metadata: its initial archive invocation exited 1 for that setup limitation, then passed in the original checkout at the same pinned HEAD with no tracked changes. Existing untracked files are excluded by its tracked-file scan. +Host `/opt/homebrew/bin/python3` is Python 3.15.2, outside a virtual environment. Docs and unit checks ran from the pinned archive `/pr60-rev-r2-f3964dd`. The full manifest check needs Git metadata: its initial archive invocation exited 1 for that setup limitation, then passed in the original checkout at the same pinned HEAD with no tracked changes. Existing untracked files are excluded by its tracked-file scan. Commands, run from the indicated root: @@ -51,7 +51,7 @@ checks[4]{command,exit,result}: tools_tests,0,"93 tests; 1 skipped; 1.143 seconds" ``` -Logs are `/private/tmp/pr60-r2-check-0.log` through `/private/tmp/pr60-r2-check-3.log` in command order. Additional behavioral checks passed. No live acceptance stack, long-duration wait or Trivy scan was run. No repository source was edited, committed or pushed; no merge, approval or closure was performed. Remaining work is the coordinator's final PR decision; no revision is requested in this review scope. +Logs are `/pr60-r2-check-0.log` through `/pr60-r2-check-3.log` in command order. Additional behavioral checks passed. No live acceptance stack, long-duration wait or Trivy scan was run. No repository source was edited, committed or pushed; no merge, approval or closure was performed. Remaining work is the coordinator's final PR decision; no revision is requested in this review scope. --- @@ -74,7 +74,7 @@ Logs are `/private/tmp/pr60-r2-check-0.log` through `/private/tmp/pr60-r2-check- ### 検証と限界 -仮想環境外のホスト `/opt/homebrew/bin/python3`(Python 3.15.2)を使用した。docs と単体試験は指定コミットを展開した `/private/tmp/pr60-rev-r2-f3964dd` で実行した。manifest 全検査は Git 情報が必要なため、最初のアーカイブ実行は環境上の理由で終了 1 となり、同じ指定 HEAD・追跡ファイルの変更なしの元 checkout で再実行して合格した。既存の未追跡ファイルは追跡ファイル限定走査の対象外である。 +仮想環境外のホスト `/opt/homebrew/bin/python3`(Python 3.15.2)を使用した。docs と単体試験は指定コミットを展開した `/pr60-rev-r2-f3964dd` で実行した。manifest 全検査は Git 情報が必要なため、最初のアーカイブ実行は環境上の理由で終了 1 となり、同じ指定 HEAD・追跡ファイルの変更なしの元 checkout で再実行して合格した。既存の未追跡ファイルは追跡ファイル限定走査の対象外である。 該当ルートから実行したコマンド: @@ -95,4 +95,4 @@ checks[4]{command,exit,result}: tools_tests,0,"93 件; 1 skip; 1.143 秒" ``` -ログは実行順に `/private/tmp/pr60-r2-check-0.log`〜`/private/tmp/pr60-r2-check-3.log`。補足の動作検証も合格した。実スタックの受入、長時間の実待機、Trivy の実走査は行っていない。リポジトリのソース変更・commit・push、merge・approve・close は行っていない。残作業は coordinator の PR 最終判断であり、このレビュー範囲で修正要求はない。 +ログは実行順に `/pr60-r2-check-0.log`〜`/pr60-r2-check-3.log`。補足の動作検証も合格した。実スタックの受入、長時間の実待機、Trivy の実走査は行っていない。リポジトリのソース変更・commit・push、merge・approve・close は行っていない。残作業は coordinator の PR 最終判断であり、このレビュー範囲で修正要求はない。 diff --git a/docs/reviews/2026-11-13_RDM-developer-guide_PR61_M2_Close_ADR_Approval_r1.md b/docs/reviews/2026-11-13_RDM-developer-guide_PR61_M2_Close_ADR_Approval_r1.md index 66008a35..d6c5ef22 100644 --- a/docs/reviews/2026-11-13_RDM-developer-guide_PR61_M2_Close_ADR_Approval_r1.md +++ b/docs/reviews/2026-11-13_RDM-developer-guide_PR61_M2_Close_ADR_Approval_r1.md @@ -7,8 +7,8 @@ - Version / 版: 0.1 — 2026-11-13. **Learned:** a milestone closure can accept missing evidence without turning an in-place upgrade into a proven one-command installation; the status board must retain that distinction. / 節目の完了は証拠の不足を受け入れられるが、既存環境の更新を一コマンド導入の証明には変えない。進捗ボードにもその区別が必要と分かった。 - Applies to / 対象: RCOSDP/RDM-developer-guide PR #61, head `dbbd288a2aacb16d0710d3e1abed51fdd6b6d85d`, base `develop` (`fb9ccdc`). REV, Codex Astra, round 1. -- Parent index / 親索引: [Reviews](../../../../../rdm-pr61-rev-r1/docs/reviews/README.md) -- Related / 関連: [SPEC-03](../../../../../rdm-pr61-rev-r1/docs/spec/03_security_gate.md), [SPEC-06](../../../../../rdm-pr61-rev-r1/docs/spec/06_documentation_conventions.md), [status board](../../../../../rdm-pr61-rev-r1/docs/status/README.md), [DEMO-20](../../../../../rdm-pr61-rev-r1/docs/demo/20_core_install_mac_studio.md). +- Parent index / 親索引: [Reviews](README.md) +- Related / 関連: [SPEC-03](../spec/03_security_gate.md), [SPEC-06](../spec/06_documentation_conventions.md), [status board](../status/README.md), [DEMO-20](../demo/20_core_install_mac_studio.md). - Used by / 利用先: PR #61 coordinator and author / PR #61 の調整担当と作成者。 --- @@ -20,7 +20,7 @@ ### Verdict and scope -**CHANGES REQUESTED**, for R1-01 and R1-02. These are documentation corrections, not a request to reverse the owner's decisions or rerun hardware tests before closing M2. I reviewed all ten changed files in a clean detached clone at `/private/tmp/rdm-pr61-rev-r1`, repository evidence and relevant GitHub PR metadata. The supplied task establishes the owner's 2026-11-13 decisions; I did not independently authenticate the conversation. No source was changed, committed or pushed; no merge, approval or closure was performed. No existing runtime or `.env` was accessed; unit tests use their own temporary synthetic fixtures. No live Docker installation or scanner run was attempted. +**CHANGES REQUESTED**, for R1-01 and R1-02. These are documentation corrections, not a request to reverse the owner's decisions or rerun hardware tests before closing M2. I reviewed all ten changed files in a clean detached clone at `/rdm-pr61-rev-r1`, repository evidence and relevant GitHub PR metadata. The supplied task establishes the owner's 2026-11-13 decisions; I did not independently authenticate the conversation. No source was changed, committed or pushed; no merge, approval or closure was performed. No existing runtime or `.env` was accessed; unit tests use their own temporary synthetic fixtures. No live Docker installation or scanner run was attempted. ### Findings @@ -75,7 +75,7 @@ checks[5]{command,result,exit}: ### 判定と範囲 -**修正要求(CHANGES REQUESTED)**。R1-01・R1-02 は文書の訂正であり、所有者の判断の撤回や M2 完了前の実機再試験を要求しない。`/private/tmp/rdm-pr61-rev-r1` の指定 head のクリーンクローンで変更 10 ファイル、リポジトリの証跡、関連 PR の情報を確認した。2026-11-13 の所有者判断は依頼の記載を根拠とし、元の会話を独立に認証してはいない。ソース変更・commit・push・merge・approve・close は行っていない。既存 runtime と `.env` はアクセスせず、単体試験は一時的な模擬データを使った。実環境の導入やスキャナー実行は行っていない。 +**修正要求(CHANGES REQUESTED)**。R1-01・R1-02 は文書の訂正であり、所有者の判断の撤回や M2 完了前の実機再試験を要求しない。`/rdm-pr61-rev-r1` の指定 head のクリーンクローンで変更 10 ファイル、リポジトリの証跡、関連 PR の情報を確認した。2026-11-13 の所有者判断は依頼の記載を根拠とし、元の会話を独立に認証してはいない。ソース変更・commit・push・merge・approve・close は行っていない。既存 runtime と `.env` はアクセスせず、単体試験は一時的な模擬データを使った。実環境の導入やスキャナー実行は行っていない。 ### 指摘 diff --git a/docs/reviews/2026-11-13_RDM-developer-guide_PR61_M2_Close_ADR_Approval_r2.md b/docs/reviews/2026-11-13_RDM-developer-guide_PR61_M2_Close_ADR_Approval_r2.md index 66800734..f153bea9 100644 --- a/docs/reviews/2026-11-13_RDM-developer-guide_PR61_M2_Close_ADR_Approval_r2.md +++ b/docs/reviews/2026-11-13_RDM-developer-guide_PR61_M2_Close_ADR_Approval_r2.md @@ -7,8 +7,8 @@ - Version / 版: 0.1 — 2026-11-13. **Learned:** distinguishing installation history from final-pin updates, and merged fixes from deployed-screen evidence, preserves the limits of milestone evidence. / 導入履歴と最終固定版への更新、修正のマージと配備後の実画面の証拠を分けることで、節目の証拠の限界を保てると分かった。 - Applies to / 対象: RCOSDP/RDM-developer-guide PR #61; REV(担当レビュワ, Codex Astra), r2; base `develop`; head `cb8ad3e731d275e2c1d3f9403289f48a4f4f8f0d`. -- Parent index / 親索引: [Reviews](../../../../../../..docs/reviews/README.md) -- Related / 関連: [SPEC-06](../../../../../../..docs/spec/06_documentation_conventions.md) · [Status](../../../../../../..docs/status/README.md) · [DEMO-20](../../../../../../..docs/demo/20_core_install_mac_studio.md) · [DEMO-21](../../../../../../..docs/demo/21_core_install_ubuntu_arm64_lima.md) +- Parent index / 親索引: [Reviews](README.md) +- Related / 関連: [SPEC-06](../spec/06_documentation_conventions.md) · [Status](../status/README.md) · [DEMO-20](../demo/20_core_install_mac_studio.md) · [DEMO-21](../demo/21_core_install_ubuntu_arm64_lima.md) - Used by / 利用先: PR #61 coordinator and author / PR #61 の調整担当と作成者。 --- diff --git a/docs/reviews/2026-11-14_RDM-developer-guide_PR62_Amd64_Final_Pins_r1.md b/docs/reviews/2026-11-14_RDM-developer-guide_PR62_Amd64_Final_Pins_r1.md index 26885218..632fdb57 100644 --- a/docs/reviews/2026-11-14_RDM-developer-guide_PR62_Amd64_Final_Pins_r1.md +++ b/docs/reviews/2026-11-14_RDM-developer-guide_PR62_Amd64_Final_Pins_r1.md @@ -27,7 +27,7 @@ 2. **R1-02 / P2 — Keep ZAP's root cause unresolved everywhere.** DEMO-22 line 8 labels the ZAP failure “environment”; the final-runs table at line 229 repeats “transient, environment” (Japanese line 490 likewise). Yet the run-4 paragraph explicitly says no ZAP output was captured and the cause is unresolved. A successful standalone rerun establishes non-recurrence, not whether the original failure was environment, installer invocation, or scanner behavior. Change both language summaries/table cells to “not reproduced; cause unresolved”; retain the admin-readiness explanation only as an untested hypothesis. The status rows already avoid that causal assertion. -3. **R1-03 / P2 — Sanitize local metadata in the added JUnit evidence.** `docs/demo/runs/acc-20261113T192132Z-1cf20d/junit.xml:1` contains ten absolute checkout paths rooted at `/home/grdm/rdm-final/` in skip messages and the VM hostname attribute. This fails the requested absence of local scratch paths/identities even though no personal credentials were found. Replace path prefixes with repository-relative paths and remove/generalize hostname, preserving counts, outcomes and durations. The pytest.log already uses repository-relative skip paths. No secrets, personal email addresses, or coordinator scratch-directory paths were found in the added evidence. +3. **R1-03 / P2 — Sanitize local metadata in the added JUnit evidence.** `docs/demo/runs/acc-20261113T192132Z-1cf20d/junit.xml:1` contains ten absolute checkout paths rooted at `/rdm-final/` in skip messages and the VM hostname attribute. This fails the requested absence of local scratch paths/identities even though no personal credentials were found. Replace path prefixes with repository-relative paths and remove/generalize hostname, preserving counts, outcomes and durations. The pytest.log already uses repository-relative skip paths. No secrets, personal email addresses, or coordinator scratch-directory paths were found in the added evidence. ### Additional corrections @@ -72,7 +72,7 @@ The clone remained clean after validation, and the remote PR head still matched 2. **R1-02 / P2 — ZAP の原因未解決を全箇所で守る。** DEMO-22 の8行目と実行表229/490行目は ZAP を「環境」と断定する。一方、第4回の本文は ZAP 自身の出力を取得せず原因未解決と明記する。再実行の成功から分かるのは再現しなかったことまでで、環境・呼び出し側・スキャナーのどれかは決められない。両言語のヘッダと表を「再現せず、原因未解決」にし、admin の準備遅れは未検証の仮説として残す。status の行は既に原因を断定していない。 -3. **R1-03 / P2 — JUnit 証跡のローカル情報を除く。** 追加した `junit.xml:1` に `/home/grdm/rdm-final/` 以下の絶対パスがスキップ理由として10件と VM hostname が残る。個人の資格情報は見つからないが、依頼されたローカル作業パス・識別情報の除去を満たさない。パスはリポジトリ相対にし、hostname は削除または一般化する。件数・結果・時間は保持する。pytest.log のスキップ理由は既に相対パスである。追加証跡に秘密、個人メール、調整担当の scratch ディレクトリは見つからなかった。 +3. **R1-03 / P2 — JUnit 証跡のローカル情報を除く。** 追加した `junit.xml:1` に `/rdm-final/` 以下の絶対パスがスキップ理由として10件と VM hostname が残る。個人の資格情報は見つからないが、依頼されたローカル作業パス・識別情報の除去を満たさない。パスはリポジトリ相対にし、hostname は削除または一般化する。件数・結果・時間は保持する。pytest.log のスキップ理由は既に相対パスである。追加証跡に秘密、個人メール、調整担当の scratch ディレクトリは見つからなかった。 ### その他の修正 diff --git a/docs/reviews/2026-11-14_RDM-developer-guide_PR62_Amd64_Final_Pins_r2.md b/docs/reviews/2026-11-14_RDM-developer-guide_PR62_Amd64_Final_Pins_r2.md index 9d26c2a5..c4580c1e 100644 --- a/docs/reviews/2026-11-14_RDM-developer-guide_PR62_Amd64_Final_Pins_r2.md +++ b/docs/reviews/2026-11-14_RDM-developer-guide_PR62_Amd64_Final_Pins_r2.md @@ -19,7 +19,7 @@ ### Verdict and scope -**PASS WITH NITS.** No blocking finding in the five changed files. GitHub's head matched the requested SHA. The shared checkout was on another commit, so checks used an exact `git archive` export at `/private/tmp/pr62-rev-r2-dddfa9a`, with a separate temporary Git index for tracked-file checks; the repository's working tree and index were not changed. +**PASS WITH NITS.** No blocking finding in the five changed files. GitHub's head matched the requested SHA. The shared checkout was on another commit, so checks used an exact `git archive` export at `/pr62-rev-r2-dddfa9a`, with a separate temporary Git index for tracked-file checks; the repository's working tree and index were not changed. ### Findings and closure @@ -38,7 +38,7 @@ On macOS with Python 3.15.2, at the exact requested head: - `PYTHONPATH=installer python3 -m unittest discover -s installer/tests`: OK, 694 tests in 64.967 s, 10 skipped. A fork deprecation warning and local port-listener warning did not fail the suite. - `git diff --check bfc8391..dddfa9adc3c48f925c46b81f1f0d684d9640c246`: PASS. -Logs: `/private/tmp/pr62-r2-docs.log`, `/private/tmp/pr62-r2-tests.log`. +Logs: `/pr62-r2-docs.log`, `/pr62-r2-tests.log`. No VM restart, acceptance rerun, scanner rerun, merge, approval, push or commit was performed. Other committed run folders containing hostnames or absolute paths are expressly outside this review and belong to the coordinator's separate PR. The only remaining recommendation here is the Japanese version-note correction. @@ -51,7 +51,7 @@ No VM restart, acceptance rerun, scanner rerun, merge, approval, push or commit ### 判定と対象 -**PASS WITH NITS(軽微な指摘あり、阻害なし)。** 変更5ファイルに阻害指摘はない。GitHubの先端は指定SHAと一致した。共有チェックアウトは別コミットだったため、指定コミットを `git archive` で `/private/tmp/pr62-rev-r2-dddfa9a` に展開し、追跡ファイルの検査には独立した一時Gitインデックスを使った。リポジトリの作業ツリーとインデックスは変更していない。 +**PASS WITH NITS(軽微な指摘あり、阻害なし)。** 変更5ファイルに阻害指摘はない。GitHubの先端は指定SHAと一致した。共有チェックアウトは別コミットだったため、指定コミットを `git archive` で `/pr62-rev-r2-dddfa9a` に展開し、追跡ファイルの検査には独立した一時Gitインデックスを使った。リポジトリの作業ツリーとインデックスは変更していない。 ### 指摘と対応確認 @@ -70,6 +70,6 @@ macOS・Python 3.15.2で指定先端の内容を検証した。 - `PYTHONPATH=installer python3 -m unittest discover -s installer/tests`: OK、694件、64.967秒、スキップ10件。fork非推奨とローカルポート待受の警告は試験失敗を生じなかった。 - `git diff --check bfc8391..dddfa9adc3c48f925c46b81f1f0d684d9640c246`: 合格。 -ログ: `/private/tmp/pr62-r2-docs.log`、`/private/tmp/pr62-r2-tests.log`。 +ログ: `/pr62-r2-docs.log`、`/pr62-r2-tests.log`。 VM再起動、受入試験・走査の再実行、マージ、承認、push、commitは行っていない。他のコミット済み実行フォルダーのhostname・絶対パスは明示的に対象外で、調整担当の別PRで扱う。今回残る推奨対応は日本語の版履歴の修正のみ。 diff --git a/docs/reviews/2026-11-14_RDM-developer-guide_PR63_Scrub_Host_Identity_r1.md b/docs/reviews/2026-11-14_RDM-developer-guide_PR63_Scrub_Host_Identity_r1.md index 00e9a23f..5a0d96dd 100644 --- a/docs/reviews/2026-11-14_RDM-developer-guide_PR63_Scrub_Host_Identity_r1.md +++ b/docs/reviews/2026-11-14_RDM-developer-guide_PR63_Scrub_Host_Identity_r1.md @@ -36,7 +36,7 @@ All **23**, not 22, changed run files matched an independent exact identity-only DEMO-21 changes only the home-path example in both languages. SPEC-04 is version 0.12, English then Japanese, with corresponding new rules and the prior 0.11 and older history preserved. The stated file count in the PR description should be corrected to 23 (20 XML and 3 log files). -From `/private/tmp/rdm-pr63-rev-r1`: +From `/rdm-pr63-rev-r1`: ```sh python3 tools/docs_check.py @@ -55,7 +55,7 @@ checks[5]{check,exit,result}: diff whitespace check, 0, no findings ``` -Audit script: `/private/tmp/pr63-audit.py`; output: `/private/tmp/pr63-audit.log`. Additional probes confirmed escaped-home leakage, resolved-root replacement, and existing secret masking. The audit script's broad identity-match list includes application URLs; focused searches and manual inspection ruled those out. The initial installer invocation omitted `PYTHONPATH` and failed import discovery (78 tests, 21 errors); the corrected command is the authoritative suite result. +Audit script: `/pr63-audit.py`; output: `/pr63-audit.log`. Additional probes confirmed escaped-home leakage, resolved-root replacement, and existing secret masking. The audit script's broad identity-match list includes application URLs; focused searches and manual inspection ruled those out. The initial installer invocation omitted `PYTHONPATH` and failed import discovery (78 tests, 21 errors); the corrected command is the authoritative suite result. ### Every changed evidence file diff --git a/docs/reviews/2026-11-14_RDM-developer-guide_PR63_Scrub_Host_Identity_r2.md b/docs/reviews/2026-11-14_RDM-developer-guide_PR63_Scrub_Host_Identity_r2.md index c59abbdd..d6eea74a 100644 --- a/docs/reviews/2026-11-14_RDM-developer-guide_PR63_Scrub_Host_Identity_r2.md +++ b/docs/reviews/2026-11-14_RDM-developer-guide_PR63_Scrub_Host_Identity_r2.md @@ -19,7 +19,7 @@ ### Verdict and scope -**Request changes: three P2 findings.** Reviewed only the requested three-file delta; historical evidence was not re-audited. GitHub head matched the assigned SHA and the PR body now says 23 files. The shared checkout had another HEAD, so execution used a `git archive` export of the target at `/private/tmp/pr63-rev-r2-4416583`. No source edits, commit, push, approval, merge or close occurred. Exactly one Japanese PR comment was posted. +**Request changes: three P2 findings.** Reviewed only the requested three-file delta; historical evidence was not re-audited. GitHub head matched the assigned SHA and the PR body now says 23 files. The shared checkout had another HEAD, so execution used a `git archive` export of the target at `/pr63-rev-r2-4416583`. No source edits, commit, push, approval, merge or close occurred. Exactly one Japanese PR comment was posted. ### R2-01 — Numeric hostname collisions corrupt evidence @@ -51,7 +51,7 @@ checks[3]{check,result}: delta whitespace, passed ``` -Logs: `/private/tmp/pr63-r2-unittests.log` and `/private/tmp/pr63-r2-docs.log`. Synthetic probes confirmed all three findings without reading operational identity files. SPEC-04 English and Japanese match and retain version 0.12, but the implemented behavior does not fully meet the encoded-path and evidence-preservation rules. The previous plain-word examples and new ampersand/backslash examples pass. No live acceptance environment was exercised. Fixes and subsequent review remain with the author/coordinator. +Logs: `/pr63-r2-unittests.log` and `/pr63-r2-docs.log`. Synthetic probes confirmed all three findings without reading operational identity files. SPEC-04 English and Japanese match and retain version 0.12, but the implemented behavior does not fully meet the encoded-path and evidence-preservation rules. The previous plain-word examples and new ampersand/backslash examples pass. No live acceptance environment was exercised. Fixes and subsequent review remain with the author/coordinator. --- @@ -62,7 +62,7 @@ Logs: `/private/tmp/pr63-r2-unittests.log` and `/private/tmp/pr63-r2-docs.log`. ### 判定と範囲 -**要修正:P2 が3件。** 指定された3ファイルの差分だけを確認し、過去の証跡の再監査は行っていない。GitHub の head は指定 SHA と一致し、PR 本文は23ファイルに修正されている。共有作業ツリーの HEAD が異なるため、対象を `git archive` で `/private/tmp/pr63-rev-r2-4416583` に書き出して検証した。ソース変更、コミット、push、承認、マージ、close は行っていない。日本語の PR コメントを1件だけ投稿した。 +**要修正:P2 が3件。** 指定された3ファイルの差分だけを確認し、過去の証跡の再監査は行っていない。GitHub の head は指定 SHA と一致し、PR 本文は23ファイルに修正されている。共有作業ツリーの HEAD が異なるため、対象を `git archive` で `/pr63-rev-r2-4416583` に書き出して検証した。ソース変更、コミット、push、承認、マージ、close は行っていない。日本語の PR コメントを1件だけ投稿した。 ### R2-01 — 数値ホスト名との衝突で証跡が壊れる @@ -94,4 +94,4 @@ checks[3]{check,result}: 差分空白検査, 成功 ``` -ログは `/private/tmp/pr63-r2-unittests.log` と `/private/tmp/pr63-r2-docs.log`。運用上の識別情報ファイルを読まず、合成入力で3件を確認した。SPEC-04 の英日は対応し版0.12を保持しているが、実装はエスケープ済みパスの除去と証跡保持の規則を完全には満たさない。従来の一般語の例と新しいアンパサンド・バックスラッシュの例は合格している。実環境の受入試験は実施していない。修正と次回レビューは作者・コーディネーターに残す。 +ログは `/pr63-r2-unittests.log` と `/pr63-r2-docs.log`。運用上の識別情報ファイルを読まず、合成入力で3件を確認した。SPEC-04 の英日は対応し版0.12を保持しているが、実装はエスケープ済みパスの除去と証跡保持の規則を完全には満たさない。従来の一般語の例と新しいアンパサンド・バックスラッシュの例は合格している。実環境の受入試験は実施していない。修正と次回レビューは作者・コーディネーターに残す。 diff --git a/docs/reviews/2026-11-14_RDM-developer-guide_PR64_Install_Volume_Guard_r1.md b/docs/reviews/2026-11-14_RDM-developer-guide_PR64_Install_Volume_Guard_r1.md index 0e8149fc..3428d2dc 100644 --- a/docs/reviews/2026-11-14_RDM-developer-guide_PR64_Install_Volume_Guard_r1.md +++ b/docs/reviews/2026-11-14_RDM-developer-guide_PR64_Install_Volume_Guard_r1.md @@ -33,7 +33,7 @@ At `installer/tests/test_install.py:75–88`, also assert the emitted report's f ### Verification -Verification used a clean `git archive` of the exact head in `/private/tmp/pr64-rev-2622bf0`, because the shared working directory was at a different commit. System Python is `/usr/bin/python3` 3.9.6. Commands ran from that archive: +Verification used a clean `git archive` of the exact head in `/pr64-rev-2622bf0`, because the shared working directory was at a different commit. System Python is `/usr/bin/python3` 3.9.6. Commands ran from that archive: ```sh PYTHONPATH=installer /usr/bin/python3 -m unittest discover -s installer/tests @@ -67,7 +67,7 @@ Compose ファイルの欠落・読み取り不能時は Docker を呼ばず空 ### 検証 -共有作業ディレクトリが別コミットだったため、指定 head の `git archive` を `/private/tmp/pr64-rev-2622bf0` に展開して検証した。システム Python は `/usr/bin/python3` 3.9.6。英語節の二つのコマンドを展開先で実行した。 +共有作業ディレクトリが別コミットだったため、指定 head の `git archive` を `/pr64-rev-2622bf0` に展開して検証した。システム Python は `/usr/bin/python3` 3.9.6。英語節の二つのコマンドを展開先で実行した。 installer 単体テストは 704 件、74.180 秒で合格(10 件 skip)。 diff --git a/docs/reviews/2026-11-14_RDM-developer-guide_PR65_Workflow_Scope_CI_r1.md b/docs/reviews/2026-11-14_RDM-developer-guide_PR65_Workflow_Scope_CI_r1.md index 80bf6a68..d7cf0e73 100644 --- a/docs/reviews/2026-11-14_RDM-developer-guide_PR65_Workflow_Scope_CI_r1.md +++ b/docs/reviews/2026-11-14_RDM-developer-guide_PR65_Workflow_Scope_CI_r1.md @@ -7,8 +7,8 @@ - Version / 版: 0.1 — 2026-11-14. **Learned:** enabling a deferred workflow exposes existing lint debt, and polling an expected status does not assert that status after exhaustion. / 保留したワークフローを有効にすると既存の lint 違反が表面化し、期待する状態を待つだけでは待機終了後にその状態を保証できない。 - Applies to: RCOSDP/RDM-developer-guide PR #65, head `8ad5238d8d43e6c77e183f367af805f3aae094a6`, base `f005512dd059ea2d9077c36d41326d0b9b5b13af`, review round 1. -- Related: [PR #65](https://github.com/RCOSDP/RDM-developer-guide/pull/65) · [SPEC-06](../wt-wf/docs/spec/06_documentation_conventions.md) · [SPEC-02](../wt-wf/docs/spec/02_installer_cli.md) · [ADR-08](../wt-wf/docs/decisions/ADR-08_Per_OS_Installer_Entry_Points.md) · [review index](../wt-wf/docs/reviews/README.md). -- Used by: the coordinator's correction decision and [status board](../wt-wf/docs/status/README.md). +- Related: [PR #65](https://github.com/RCOSDP/RDM-developer-guide/pull/65) · [SPEC-06](../spec/06_documentation_conventions.md) · [SPEC-02](../spec/02_installer_cli.md) · [ADR-08](../decisions/ADR-08_Per_OS_Installer_Entry_Points.md) · [review index](README.md). +- Used by: the coordinator's correction decision and [status board](../status/README.md). --- diff --git a/docs/reviews/2026-11-14_RDM-developer-guide_PR65_Workflow_Scope_CI_r2.md b/docs/reviews/2026-11-14_RDM-developer-guide_PR65_Workflow_Scope_CI_r2.md index 19d22a47..22d298ee 100644 --- a/docs/reviews/2026-11-14_RDM-developer-guide_PR65_Workflow_Scope_CI_r2.md +++ b/docs/reviews/2026-11-14_RDM-developer-guide_PR65_Workflow_Scope_CI_r2.md @@ -7,8 +7,8 @@ - Version / 版: 0.1 — 2026-11-14. **Learned:** an explicit status assertion, a separate readiness wait, and an EXIT trap make smoke-test failures observable and cleanable. / 明示的な状態検証、独立した準備待ち、EXIT trap により、スモーク試験の失敗を観測して後片付けできると分かった。 - Applies to: RCOSDP/RDM-developer-guide PR #65, head `c7f8a54170b961b3bf5d77f68bbb64113270cbf7`, base branch `develop`, round 2. -- Related: [r2 PR comment](https://github.com/RCOSDP/RDM-developer-guide/pull/65#issuecomment-6872792550) · [PR #65](https://github.com/RCOSDP/RDM-developer-guide/pull/65) · [r1](2026-11-14_RDM-developer-guide_PR65_Workflow_Scope_CI_r1.md) · [SPEC-06](../wt-wf/docs/spec/06_documentation_conventions.md) · [review index](../wt-wf/docs/reviews/README.md). -- Used by: coordinator's integration decision and [status board](../wt-wf/docs/status/README.md). +- Related: [r2 PR comment](https://github.com/RCOSDP/RDM-developer-guide/pull/65#issuecomment-6872792550) · [PR #65](https://github.com/RCOSDP/RDM-developer-guide/pull/65) · [r1](2026-11-14_RDM-developer-guide_PR65_Workflow_Scope_CI_r1.md) · [SPEC-06](../spec/06_documentation_conventions.md) · [review index](README.md). +- Used by: coordinator's integration decision and [status board](../status/README.md). --- diff --git a/docs/reviews/2026-11-14_RDM-developer-guide_PR68_Status_Publication_r1.md b/docs/reviews/2026-11-14_RDM-developer-guide_PR68_Status_Publication_r1.md index a49f20a7..23638801 100644 --- a/docs/reviews/2026-11-14_RDM-developer-guide_PR68_Status_Publication_r1.md +++ b/docs/reviews/2026-11-14_RDM-developer-guide_PR68_Status_Publication_r1.md @@ -7,7 +7,7 @@ - Version / 版: 0.1 — 2026-11-14. **Learned:** publication records can be verified against merge metadata and included commits together. / 公開記録はマージ情報と収録コミットを組み合わせて検証できる。 - Applies to: RCOSDP/RDM-developer-guide PR #68, head `bc7e1aa6e2e55144f38a9f7da19eefbf5742977a`, base `develop`; REV(担当レビュワ, Codex Astra). -- Related: [Review index](../wt-pub/docs/reviews/README.md) · [SPEC-06](../wt-pub/docs/spec/06_documentation_conventions.md) · [Status board](../wt-pub/docs/status/README.md) · [PR #68](https://github.com/RCOSDP/RDM-developer-guide/pull/68). +- Related: [Review index](README.md) · [SPEC-06](../spec/06_documentation_conventions.md) · [Status board](../status/README.md) · [PR #68](https://github.com/RCOSDP/RDM-developer-guide/pull/68). - Used by: the author and coordinating reviewer. --- diff --git a/docs/reviews/2026-11-14_RDM-developer-guide_PR69_DEMO20_macOS_Fresh_r1.md b/docs/reviews/2026-11-14_RDM-developer-guide_PR69_DEMO20_macOS_Fresh_r1.md index 24cd69a0..fdc889ed 100644 --- a/docs/reviews/2026-11-14_RDM-developer-guide_PR69_DEMO20_macOS_Fresh_r1.md +++ b/docs/reviews/2026-11-14_RDM-developer-guide_PR69_DEMO20_macOS_Fresh_r1.md @@ -7,7 +7,7 @@ - Version / 版: 0.1 — 2026-11-14. **Learned:** report timestamps and preserved run files substantiate the install outcome, while current Docker state cannot independently prove every historical preparation detail. / レポート時刻と保存された実行ファイルは導入結果を裏づけるが、現在の Docker 状態だけでは過去の準備の細部まで独立に証明できない。 - Applies to: RCOSDP/RDM-developer-guide PR #69, head `acc057ebd6c5ed5430ba7c742c64d82693b9c1e3`, base `develop`; REV round 1. -- Related: [PR #69](https://github.com/RCOSDP/RDM-developer-guide/pull/69) · [Review index](../wt-demo20-macfresh/docs/reviews/README.md) · [SPEC-06](../wt-demo20-macfresh/docs/spec/06_documentation_conventions.md) · [DEMO-20](../wt-demo20-macfresh/docs/demo/20_core_install_mac_studio.md) · [Status](../wt-demo20-macfresh/docs/status/README.md) +- Related: [PR #69](https://github.com/RCOSDP/RDM-developer-guide/pull/69) · [Review index](README.md) · [SPEC-06](../spec/06_documentation_conventions.md) · [DEMO-20](../demo/20_core_install_mac_studio.md) · [Status](../status/README.md) --- diff --git a/docs/reviews/2026-11-14_RDM-ember-osf-web-dev2027_PR5_Node24_CI_r1.md b/docs/reviews/2026-11-14_RDM-ember-osf-web-dev2027_PR5_Node24_CI_r1.md index ec1d13a8..f4e9cf30 100644 --- a/docs/reviews/2026-11-14_RDM-ember-osf-web-dev2027_PR5_Node24_CI_r1.md +++ b/docs/reviews/2026-11-14_RDM-ember-osf-web-dev2027_PR5_Node24_CI_r1.md @@ -7,7 +7,7 @@ - Version / 版: 0.1 — 2026-11-14. **Learned:** Selecting Node for project commands does not change the JavaScript runtime declared by an Action. / プロジェクト用 Node の選択は Action 自身が宣言する実行環境を変更しない。 - Applies to: RCOSDP/RDM-ember-osf-web-dev2027 PR #5, round 1; REV(担当レビュワ, Codex Astra). -- Related: [Review index](../../../../../../..docs/reviews/README.md); [SPEC-06](../../../../../../..docs/spec/06_documentation_conventions.md). +- Related: [Review index](README.md); [SPEC-06](../spec/06_documentation_conventions.md). - Used by: coordinator's PR #5 disposition; this scratch record is not committed. --- diff --git a/docs/reviews/2026-11-14_RDM-osf.io-dev2027_PR22_Private_Security_CI_r1.md b/docs/reviews/2026-11-14_RDM-osf.io-dev2027_PR22_Private_Security_CI_r1.md index 786d931a..48cb6946 100644 --- a/docs/reviews/2026-11-14_RDM-osf.io-dev2027_PR22_Private_Security_CI_r1.md +++ b/docs/reviews/2026-11-14_RDM-osf.io-dev2027_PR22_Private_Security_CI_r1.md @@ -8,7 +8,7 @@ - Version / 版: 0.1 — 2026-11-14. **Learned:** an implemented test task provides no continuous coverage until the workflow selects it; enabling the existing private task collects three otherwise omitted files without changing public-mode jobs. / 実装済みの試験タスクもワークフローから選ばれなければ継続検査にならず、既存の非公開タスクを有効にすると公開モードのジョブを変えずに未収集3ファイルを検査できる。 - Applies to: RCOSDP/RDM-osf.io-dev2027 PR #22, branch `ci/private-security-matrix`, head `3aad270c72577b273b52152a07628b60f0db687c`, base `develop` at `90b10661c8dcf039c18fdd214aef98191f97497c`. - Reviewer: REV(担当レビュワ, Codex Astra). -- Related: [Review index](README.md) · [SPEC-06](https://github.com/RCOSDP/RDM-developer-guide/blob/f005512dd059ea2d9077c36d41326d0b9b5b13af/docs/spec/06_documentation_conventions.md) · [PR #22](https://github.com/RCOSDP/RDM-osf.io-dev2027/pull/22) · [Issue #16](https://github.com/RCOSDP/RDM-osf.io-dev2027/issues/16). +- Related: [Review index](README.md) · [SPEC-06](https://github.com/RCOSDP/RDM-developer-guide/blob/f005512dd059ea2d9077c36d41326d0b9b5b13af/docs/spec/06_documentation_conventions.md) · [PR #22](RDM-osf.io-dev2027 PR #22) · [Issue #16](RDM-osf.io-dev2027 issue #16). - Used by: issue #16 completion assessment and the coordinator's merge decision. Internal review record; private repository links are intended for authorized readers. --- @@ -51,7 +51,7 @@ private_security_results[2]{event,run,job,conclusion,passed,warnings,test_second push,43562240653,132356584468,success,351,76876,587.04,2026-11-14T06:28:38Z ``` -Evidence: [PR-event job](https://github.com/RCOSDP/RDM-osf.io-dev2027/actions/runs/43562245773/job/132358487871), [push-event job](https://github.com/RCOSDP/RDM-osf.io-dev2027/actions/runs/43562240653/job/132356584468), [saved PR log](PR22_r1_private_security_pr.log), [saved push log](PR22_r1_private_security_push.log). The warning counts are reported separately from failures. No failing tests or failure cause remain to diagnose. +Evidence: [PR-event job](RDM-osf.io-dev2027 Actions run 43562245773), [push-event job](RDM-osf.io-dev2027 Actions run 43562240653), saved PR log (`PR22_r1_private_security_pr.log`, external evidence not kept in the repository), saved push log (`PR22_r1_private_security_push.log`, external evidence not kept in the repository). The warning counts are reported separately from failures. No failing tests or failure cause remain to diagnose. ### Scope and remaining work @@ -97,7 +97,7 @@ private_security_results[2]{event,run,job,conclusion,passed,warnings,test_second push,43562240653,132356584468,success,351,76876,587.04,2026-11-14T06:28:38Z ``` -証拠: [PR 起点ジョブ](https://github.com/RCOSDP/RDM-osf.io-dev2027/actions/runs/43562245773/job/132358487871)、[push 起点ジョブ](https://github.com/RCOSDP/RDM-osf.io-dev2027/actions/runs/43562240653/job/132356584468)、[保存した PR ログ](PR22_r1_private_security_pr.log)、[保存した push ログ](PR22_r1_private_security_push.log)。警告件数は失敗件数と区別して記録した。原因調査が必要な失敗試験はない。 +証拠: [PR 起点ジョブ](RDM-osf.io-dev2027 Actions run 43562245773)、[push 起点ジョブ](RDM-osf.io-dev2027 Actions run 43562240653)、保存した PR ログ (`PR22_r1_private_security_pr.log`, external evidence not kept in the repository)、保存した push ログ (`PR22_r1_private_security_push.log`, external evidence not kept in the repository)。警告件数は失敗件数と区別して記録した。原因調査が必要な失敗試験はない。 ### 範囲と残作業 diff --git a/docs/reviews/2026-11-15_RDM-developer-guide_PR75_osf_Pin_94e6817_r1.md b/docs/reviews/2026-11-15_RDM-developer-guide_PR75_osf_Pin_94e6817_r1.md index 7f7e6c1c..8dc1510e 100644 --- a/docs/reviews/2026-11-15_RDM-developer-guide_PR75_osf_Pin_94e6817_r1.md +++ b/docs/reviews/2026-11-15_RDM-developer-guide_PR75_osf_Pin_94e6817_r1.md @@ -7,7 +7,7 @@ - Version / 版: 0.1 — 2026-11-15. **Learned:** report timestamps corroborate completion times, but do not establish every command start time; acceptance on a pin does not exercise every upstream change. / レポートの時刻は終了時刻の裏付けになるが、各コマンドの開始時刻すべてを証明しない。固定版の受入合格は上流の全変更の実機試験ではない。 - Applies to: RCOSDP/RDM-developer-guide PR #75, head `ab98ea35222bbb7388ec707faa9b50a015c17ae0`, base `develop`; reviewer REV(担当レビュワ, Codex Astra). -- Related: [review index](../pr75-rev-r1/docs/reviews/README.md), [SPEC-06](../pr75-rev-r1/docs/spec/06_documentation_conventions.md), [DEMO-20](../pr75-rev-r1/docs/demo/20_core_install_mac_studio.md), [status board](../pr75-rev-r1/docs/status/README.md), [PR #75](https://github.com/RCOSDP/RDM-developer-guide/pull/75). +- Related: [review index](README.md), [SPEC-06](../spec/06_documentation_conventions.md), [DEMO-20](../demo/20_core_install_mac_studio.md), [status board](../status/README.md), [PR #75](https://github.com/RCOSDP/RDM-developer-guide/pull/75). --- diff --git a/docs/reviews/2026-11-15_RDM-developer-guide_PR76_Up_Stale_Render_r1.md b/docs/reviews/2026-11-15_RDM-developer-guide_PR76_Up_Stale_Render_r1.md index 2df7e540..04c78b34 100644 --- a/docs/reviews/2026-11-15_RDM-developer-guide_PR76_Up_Stale_Render_r1.md +++ b/docs/reviews/2026-11-15_RDM-developer-guide_PR76_Up_Stale_Render_r1.md @@ -8,8 +8,8 @@ - Version / 版: 0.1 — 2026-11-15. **Learned:** render freshness is equality with the current pin, not chronological age; unpinned build-report tags must remain usable. / renderの新鮮さは時系列の古さでなく現在の固定版との一致であり、固定版がない部品のビルド記録のタグは使い続けられなければならない。 - Applies to: [PR #76](https://github.com/RCOSDP/RDM-developer-guide/pull/76), `fix/up-stale-render` → `develop`, head `2ca87eaad6ca09960f2791823db7d8adc6c5240a`. - Reviewer: REV(担当レビュワ, Codex Astra). -- Related: [Review index](../wt-upstale/docs/reviews/README.md) · [SPEC-01](../wt-upstale/docs/spec/01_manifest.md) · [SPEC-02](../wt-upstale/docs/spec/02_installer_cli.md) · [SPEC-06](../wt-upstale/docs/spec/06_documentation_conventions.md). -- Used by: [Status board](../wt-upstale/docs/status/README.md). +- Related: [Review index](README.md) · [SPEC-01](../spec/01_manifest.md) · [SPEC-02](../spec/02_installer_cli.md) · [SPEC-06](../spec/06_documentation_conventions.md). +- Used by: [Status board](../status/README.md). --- diff --git a/docs/reviews/2026-11-15_RDM-developer-guide_PR78_S1_y_websocket_r1.md b/docs/reviews/2026-11-15_RDM-developer-guide_PR78_S1_y_websocket_r1.md index 4b02266f..dfd474a1 100644 --- a/docs/reviews/2026-11-15_RDM-developer-guide_PR78_S1_y_websocket_r1.md +++ b/docs/reviews/2026-11-15_RDM-developer-guide_PR78_S1_y_websocket_r1.md @@ -8,7 +8,7 @@ Version / 版: 0.1 — 2026-11-15. **Learned:** A loopback-only assertion can still fail a mandatory scanner when its diagnostic resembles an insecure connection. / ループバック限定の検証でも、診断文が安全でない接続に見えると必須の検査を落とす。 **Applies to:** RCOSDP/RDM-developer-guide PR #78, round 1; REV(担当レビュワ, Codex Astra). -**Related:** [Review index](../wt-s1/docs/reviews/README.md) · [SPEC-04](../wt-s1/docs/spec/04_acceptance_tests.md) · [SPEC-06](../wt-s1/docs/spec/06_documentation_conventions.md). +**Related:** [Review index](README.md) · [SPEC-04](../spec/04_acceptance_tests.md) · [SPEC-06](../spec/06_documentation_conventions.md). **Used by:** [PR #78](https://github.com/RCOSDP/RDM-developer-guide/pull/78). --- diff --git a/docs/reviews/2026-11-15_RDM-developer-guide_PR78_S1_y_websocket_r2.md b/docs/reviews/2026-11-15_RDM-developer-guide_PR78_S1_y_websocket_r2.md index a46b5415..6b9beef2 100644 --- a/docs/reviews/2026-11-15_RDM-developer-guide_PR78_S1_y_websocket_r2.md +++ b/docs/reviews/2026-11-15_RDM-developer-guide_PR78_S1_y_websocket_r2.md @@ -8,7 +8,7 @@ Version / 版: 0.1 — 2026-11-15. **Learned:** Rephrasing a diagnostic can remove a scanner false positive while preserving every executable assertion. / 診断文の言い換えで、実行する検証をすべて保ったまま検査の誤検知を解消できる。 **Applies to:** RCOSDP/RDM-developer-guide PR #78, round 2; REV(担当レビュワ, Codex Astra). -**Related:** [Review index](../wt-s1/docs/reviews/README.md) · [Round 1](2026-11-15_RDM-developer-guide_PR78_S1_y_websocket_r1.md) · [SPEC-04](../wt-s1/docs/spec/04_acceptance_tests.md) · [SPEC-06](../wt-s1/docs/spec/06_documentation_conventions.md). +**Related:** [Review index](README.md) · [Round 1](2026-11-15_RDM-developer-guide_PR78_S1_y_websocket_r1.md) · [SPEC-04](../spec/04_acceptance_tests.md) · [SPEC-06](../spec/06_documentation_conventions.md). **Used by:** [PR #78](https://github.com/RCOSDP/RDM-developer-guide/pull/78). --- diff --git a/docs/reviews/2026-11-15_RDM-developer-guide_PR82_DEMO22_Empty_Docker_r1.md b/docs/reviews/2026-11-15_RDM-developer-guide_PR82_DEMO22_Empty_Docker_r1.md index 62b047a8..6e49e655 100644 --- a/docs/reviews/2026-11-15_RDM-developer-guide_PR82_DEMO22_Empty_Docker_r1.md +++ b/docs/reviews/2026-11-15_RDM-developer-guide_PR82_DEMO22_Empty_Docker_r1.md @@ -7,7 +7,7 @@ - Version / 版: 0.1 — 2026-11-15. **Learned:** report duration and outer command timestamps have different boundaries; a successful retry supports an interpretation of a transient crash, not a proven cause. / 報告の所要時間と外側のコマンド時刻は測定範囲が違い、再試行の成功は一過性の異常という解釈を支持するが原因の証明ではない。 - Applies to: RCOSDP/RDM-developer-guide PR #82, head `ba4692c6b2e6b7d8bd8cb02c3f933b8f13b5660b`, base `078b30e82e14e7b4a573085a819646d76a904e2c`. -- Related: [PR #82](https://github.com/RCOSDP/RDM-developer-guide/pull/82) · [SPEC-06](../wt-d22b/docs/spec/06_documentation_conventions.md) · [Review index](../wt-d22b/docs/reviews/README.md) · [DEMO-22](../wt-d22b/docs/demo/22_core_install_ubuntu_amd64_lima.md) · [Status board](../wt-d22b/docs/status/README.md) +- Related: [PR #82](https://github.com/RCOSDP/RDM-developer-guide/pull/82) · [SPEC-06](../spec/06_documentation_conventions.md) · [Review index](README.md) · [DEMO-22](../demo/22_core_install_ubuntu_amd64_lima.md) · [Status board](../status/README.md) --- diff --git a/docs/reviews/2026-11-15_RDM-developer-guide_PR83_S4_OASys_r1.md b/docs/reviews/2026-11-15_RDM-developer-guide_PR83_S4_OASys_r1.md index 3ce11e0d..c5d0b5dc 100644 --- a/docs/reviews/2026-11-15_RDM-developer-guide_PR83_S4_OASys_r1.md +++ b/docs/reviews/2026-11-15_RDM-developer-guide_PR83_S4_OASys_r1.md @@ -7,7 +7,7 @@ - Version / 版: 0.1 — 2026-11-15. **Learned:** grep whole-line matching does not validate an entire multiline environment value. / grep の行全体一致は、複数行の環境変数全体を検証しない。 - Applies to: RCOSDP/RDM-developer-guide PR #83, head `2f3770bc746fe590c91ef85e737aa657ab453dab`, base `6dabbf11617b3c96ef1350b01cc7ece6a3036999`. -- Related: [Review index](../rev-pr83/docs/reviews/README.md) · [SPEC-06](../rev-pr83/docs/spec/06_documentation_conventions.md) · [SPEC-04](../rev-pr83/docs/spec/04_acceptance_tests.md) · [Status](../rev-pr83/docs/status/README.md). +- Related: [Review index](README.md) · [SPEC-06](../spec/06_documentation_conventions.md) · [SPEC-04](../spec/04_acceptance_tests.md) · [Status](../status/README.md). --- diff --git a/docs/reviews/2026-11-15_RDM-developer-guide_PR83_S4_OASys_r2.md b/docs/reviews/2026-11-15_RDM-developer-guide_PR83_S4_OASys_r2.md index 48383aca..97210123 100644 --- a/docs/reviews/2026-11-15_RDM-developer-guide_PR83_S4_OASys_r2.md +++ b/docs/reviews/2026-11-15_RDM-developer-guide_PR83_S4_OASys_r2.md @@ -7,7 +7,7 @@ - Version / 版: 0.1 — 2026-11-15. **Learned:** rejecting CR and LF before line-oriented matching closes the reproduced substitution bypass. / 行単位の照合より先に CR と LF を拒否すると、再現した置換の迂回を防げる。 - Applies to: PR #83 head `07456ad5279fdb7a5f14e8ceead08eaf887933d7`, changes since `2f3770b`. -- Related: [Review index](../rev-pr83-r2/docs/reviews/README.md) · [SPEC-06](../rev-pr83-r2/docs/spec/06_documentation_conventions.md) · [Round 1](2026-11-15_RDM-developer-guide_PR83_S4_OASys_r1.md) · [Status](../rev-pr83-r2/docs/status/README.md). +- Related: [Review index](README.md) · [SPEC-06](../spec/06_documentation_conventions.md) · [Round 1](2026-11-15_RDM-developer-guide_PR83_S4_OASys_r1.md) · [Status](../status/README.md). --- diff --git a/docs/reviews/2026-11-15_RDM-developer-guide_PR84_osf_Pin_07cc56d_r1.md b/docs/reviews/2026-11-15_RDM-developer-guide_PR84_osf_Pin_07cc56d_r1.md index b240dbd7..e9a5f9e8 100644 --- a/docs/reviews/2026-11-15_RDM-developer-guide_PR84_osf_Pin_07cc56d_r1.md +++ b/docs/reviews/2026-11-15_RDM-developer-guide_PR84_osf_Pin_07cc56d_r1.md @@ -7,7 +7,7 @@ - Version / 版: 0.1 — 2026-11-15. **Learned:** inherited Mapcore reads are constrained by the caller's read-only authorization, while the committed acceptance evidence establishes the core scenarios rather than the Mapcore behavior itself. / Mapcore の継承読み取りは呼出側の読み取り専用認可に制限され、コミットされた受入記録が示すのはコアシナリオであり Mapcore 自体の動作ではない。 - Applies to: RDM-developer-guide PR #84, head `b6ad0bd7bec7667eb54c23a0a2cf9c79a5e80d7a`, base `6dabbf1`; REV(担当レビュワ, Codex Astra), round 1. -- Related: [review index](../rev-pr84/docs/reviews/README.md) · [SPEC-06](../rev-pr84/docs/spec/06_documentation_conventions.md) · [DEMO-20](../rev-pr84/docs/demo/20_core_install_mac_studio.md) · [status board](../rev-pr84/docs/status/README.md) · [PR #84](https://github.com/RCOSDP/RDM-developer-guide/pull/84). +- Related: [review index](README.md) · [SPEC-06](../spec/06_documentation_conventions.md) · [DEMO-20](../demo/20_core_install_mac_studio.md) · [status board](../status/README.md) · [PR #84](https://github.com/RCOSDP/RDM-developer-guide/pull/84). - Used by: the coordinator's decision on PR #84. This record is outside the repository and is not committed. --- diff --git a/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r1.md b/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r1.md index be5aca79..c09d04b3 100644 --- a/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r1.md +++ b/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r1.md @@ -9,7 +9,7 @@ - Applies to / 対象: RCOSDP/RDM-cas-overlay-dev2027 PR #5, head `cb064a5ecef396a07dcfdff98adb89b0c1e7d9b7`, develop `22610b7681a29241365108e5129bcd43ade4610e`. - Posted comment / 投稿済みコメント: PR #5 issuecomment-6900903245 (one Japanese review comment / 日本語レビューコメント1件). - Reviewer / 担当: REV(担当レビュワ, Codex Astra). -- Related / 関連: [Review index](../../../../../../..docs/reviews/README.md) · [SPEC-06](../../../../../../..docs/spec/06_documentation_conventions.md). +- Related / 関連: [Review index](README.md) · [SPEC-06](../spec/06_documentation_conventions.md). - Used by / 利用先: PR #5 revision and subsequent guide external-asset exception review / PR #5 修正と後続のガイド外部アセット例外見直し。 --- @@ -47,7 +47,7 @@ The independent mock-DOM probe passed DOMContentLoaded and already-ready initial ### Evidence and verification limits -Evidence files are outside the clone: [source scan](../rev-cas5-scan.log), [mock-DOM probe](../rev-cas5-probe.cjs), [probe results](../rev-cas5-probe.log), [headless browser probe](../rev-cas5-browser.cjs), [browser results](../rev-cas5-browser.log), and [Maven build log](../rev-cas5-build.log). The probes run from the scratchpad directory with `node rev-cas5-probe.cjs` and `node rev-cas5-browser.cjs`; the latter imports the locally installed Playwright package and launches installed Chrome. All browser network requests are intercepted: only the synthetic same-origin HTML and exact cas.js bytes are fulfilled, and every other request is aborted. Monitoring fixtures intentionally omit JSP rendering and reproduce the extracted JavaScript failures. +Evidence files are outside the clone: [source scan](../../../rev-cas5-scan.log), [mock-DOM probe](../../../rev-cas5-probe.cjs), [probe results](../../../rev-cas5-probe.log), [headless browser probe](../../../rev-cas5-browser.cjs), [browser results](../../../rev-cas5-browser.log), and [Maven build log](../../../rev-cas5-build.log). The probes run from the scratchpad directory with `node rev-cas5-probe.cjs` and `node rev-cas5-browser.cjs`; the latter imports the locally installed Playwright package and launches installed Chrome. All browser network requests are intercepted: only the synthetic same-origin HTML and exact cas.js bytes are fulfilled, and every other request is aborted. Monitoring fixtures intentionally omit JSP rendering and reproduce the extracted JavaScript failures. Local Maven was absent and the host Java launcher reported no runtime. An initial bind-mounted container could not see the local POM, so the fresh checkout was streamed with tar into an isolated, disposable Maven/Temurin 8 container using the Dockerfile-pinned digest `sha256:0f402712b7a469189cfea5ec69cdcadcb32b22aaf1ea62777ab925c3d8470175`. `mvn -B -DskipTests=false package` succeeded in 1:55, with 109 OAuth tests and 41 OSF tests passing (150 total, no failures/errors/skips), and generated `/code/cas-server-webapp/target/cas.war`. The default nocheck profile still disables other static checks; this is not a claim those checks ran. The disposable container was automatically removed; the WAR was not retained or deployed. No standalone CAS runtime with its backend configuration was provisioned, so servlet-generated output remains unverified. The downloaded upstream WAR has SHA-256 `78d1f99f9eeacffeb6df0bb475490539c7509e3698cce60894a76276c1d34030`. @@ -88,7 +88,7 @@ No real server-rendered login/institution/OTP/error JSP was loaded with network ### 証拠と未検証範囲 -証拠は clone 外に保存した。[参照検索](../rev-cas5-scan.log)、[模擬 DOM 試験](../rev-cas5-probe.cjs)、[試験結果](../rev-cas5-probe.log)、[headless browser 試験](../rev-cas5-browser.cjs)、[browser 結果](../rev-cas5-browser.log)、[Maven ビルドログ](../rev-cas5-build.log)。scratchpad ディレクトリで `node rev-cas5-probe.cjs` と `node rev-cas5-browser.cjs` を実行した。後者はローカルの Playwright とインストール済み Chrome を使用する。ブラウザ通信は全件捕捉し、模擬 HTML と cas.js の実ファイルだけを同一 origin として返し、それ以外を中止する。監視画面の fixture は JSP を描画せず、抽出した JavaScript の失敗を再現する。 +証拠は clone 外に保存した。[参照検索](../../../rev-cas5-scan.log)、[模擬 DOM 試験](../../../rev-cas5-probe.cjs)、[試験結果](../../../rev-cas5-probe.log)、[headless browser 試験](../../../rev-cas5-browser.cjs)、[browser 結果](../../../rev-cas5-browser.log)、[Maven ビルドログ](../../../rev-cas5-build.log)。scratchpad ディレクトリで `node rev-cas5-probe.cjs` と `node rev-cas5-browser.cjs` を実行した。後者はローカルの Playwright とインストール済み Chrome を使用する。ブラウザ通信は全件捕捉し、模擬 HTML と cas.js の実ファイルだけを同一 origin として返し、それ以外を中止する。監視画面の fixture は JSP を描画せず、抽出した JavaScript の失敗を再現する。 ホストには Maven がなく、Java ランチャーも実行環境なしと報告した。最初の bind mount コンテナにはローカル POM が見えなかったため、fresh checkout を tar で隔離した使い捨て Maven/Temurin 8 コンテナへ転送した。イメージは Dockerfile が固定する digest `sha256:0f402712b7a469189cfea5ec69cdcadcb32b22aaf1ea62777ab925c3d8470175`。`mvn -B -DskipTests=false package` は1分55秒で成功し、OAuth 109件・OSF 41件の計150件が失敗・エラー・skip なしで合格、`/code/cas-server-webapp/target/cas.war` を生成した。既定の nocheck profile は他の静的検査を無効にするため、それらの実行成功は主張しない。使い捨てコンテナは自動削除され、生成 WAR は保存・配備していない。バックエンド設定を備えた独立 CAS 実行環境は用意しておらず、サーブレット生成結果は未検証である。取得した上流 WAR の SHA-256 は `78d1f99f9eeacffeb6df0bb475490539c7509e3698cce60894a76276c1d34030`。 diff --git a/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r2.md b/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r2.md index 1830f7a7..5a7d0b79 100644 --- a/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r2.md +++ b/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r2.md @@ -9,7 +9,7 @@ - Applies to: RDM-cas-overlay-dev2027 PR #5 `ebff4e56ff0489da6aa55d5074721e14f5bf0d1d`. - Posted comment / 投稿コメント: issuecomment-6901736021. - Reviewer: REV(担当レビュワ, Codex Astra). -- Related: [Review index](../rev-guide88/docs/reviews/README.md) · [SPEC-06](../rev-guide88/docs/spec/06_documentation_conventions.md) · [r1](2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r1.md). Used by: CAS revision and guide PR #88 integration. +- Related: [Review index](README.md) · [SPEC-06](../spec/06_documentation_conventions.md) · [r1](2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r1.md). Used by: CAS revision and guide PR #88 integration. --- @@ -53,7 +53,7 @@ No `/status/config` or `/statistics/ssosessions` dependency was found in guide d The Dockerfile-pinned Maven/Temurin 8 digest was used in an isolated container. From a tar-streamed copy of the checkout, `mvn -B -DskipTests=false package` succeeded in 1:32, with OAuth 109 and OSF 45 tests (154 total, no failures/errors/skips), and the exact shipped check accepted the built WAR: 57 served files. Other checks disabled by the default nocheck profile are not claimed. The retained WAR is `../rev-cas5-r2.war`; its static extraction is `../rev-cas5-r2-served`. Reviewer containers were removed. -The unmodified upstream CAS 4.1.5 WAR was also rejected (10 unique lines). This is a different artifact from the author's old overlay WAR; the claimed count of eight for that artifact was not independently reproduced. Evidence: [build log](../rev-cas5-r2-build.log), [static scan](../rev-cas5-r2-served-scan.log), [mutation results](../rev-cas5-r2-mutants.log), [fixture inputs](../rev-cas5-r2-mutants/cases.json), [old artifact and space-filename results](../rev-cas5-r2-old-check.log). No live JSP rendering or stack acceptance was rerun; guide evidence is reviewed separately. +The unmodified upstream CAS 4.1.5 WAR was also rejected (10 unique lines). This is a different artifact from the author's old overlay WAR; the claimed count of eight for that artifact was not independently reproduced. Evidence: [build log](../../../rev-cas5-r2-build.log), [static scan](../../../rev-cas5-r2-served-scan.log), [mutation results](../../../rev-cas5-r2-mutants.log), [fixture inputs](../../../rev-cas5-r2-mutants/cases.json), [old artifact and space-filename results](../../../rev-cas5-r2-old-check.log). No live JSP rendering or stack acceptance was rerun; guide evidence is reviewed separately. --- @@ -86,4 +86,4 @@ Google Fonts・html5shiv・Analytics の読み込みは除去され、CSS はシ Dockerfile 固定の Maven/Temurin 8 digest の隔離コンテナへ checkout を tar 転送し、`mvn -B -DskipTests=false package` が1分32秒で成功した。OAuth 109件・OSF 45件、合計154件が失敗・エラー・skip なし。出荷する検査スクリプトは生成 WAR の57ファイルを合格と判定した。既定 nocheck profile が無効化する他の検査の成功は主張しない。WAR は `../rev-cas5-r2.war`、展開結果は `../rev-cas5-r2-served` に保存し、レビュワのコンテナは削除した。 -未変更の上流 CAS 4.1.5 WAR も拒否した(重複除去後10行)。これは作者の旧 overlay WAR とは別であり、作者の8行という件数は独立再現していない。証拠は [build log](../rev-cas5-r2-build.log)、[静的検索](../rev-cas5-r2-served-scan.log)、[変異試験結果](../rev-cas5-r2-mutants.log)、[入力](../rev-cas5-r2-mutants/cases.json)、[旧成果物と空白ファイル名](../rev-cas5-r2-old-check.log)。実 JSP 描画とスタック受入は再実行せず、ガイドの証拠は別レビューで扱う。 +未変更の上流 CAS 4.1.5 WAR も拒否した(重複除去後10行)。これは作者の旧 overlay WAR とは別であり、作者の8行という件数は独立再現していない。証拠は [build log](../../../rev-cas5-r2-build.log)、[静的検索](../../../rev-cas5-r2-served-scan.log)、[変異試験結果](../../../rev-cas5-r2-mutants.log)、[入力](../../../rev-cas5-r2-mutants/cases.json)、[旧成果物と空白ファイル名](../../../rev-cas5-r2-old-check.log)。実 JSP 描画とスタック受入は再実行せず、ガイドの証拠は別レビューで扱う。 diff --git a/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r3.md b/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r3.md index 0a7269ce..3355773e 100644 --- a/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r3.md +++ b/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r3.md @@ -9,7 +9,7 @@ - Applies to: RDM-cas-overlay-dev2027 PR #5 `878099b0405060a0d05c0d52cb286e53abf381e1`. - Posted comment / 投稿コメント: issuecomment-6902351976. - Reviewer: REV(担当レビュワ, Codex Astra). -- Related: [Review index](../rev-guide88/docs/reviews/README.md) · [SPEC-06](../rev-guide88/docs/spec/06_documentation_conventions.md) · [r2](2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r2.md). Used by: CAS revision and guide PR #88 pin integration. +- Related: [Review index](README.md) · [SPEC-06](../spec/06_documentation_conventions.md) · [r2](2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r2.md). Used by: CAS revision and guide PR #88 pin integration. --- @@ -58,7 +58,7 @@ Either explicitly document intentional removal of all service logos, or provide The checkout was streamed into an isolated container using the exact Dockerfile-pinned Maven/Temurin 8 digest. `mvn -B package -DskipTests=false` succeeded in 1:31, OAuth 109 and OSF 45 tests, with zero failures/errors/skips. `sh etc/check-war-assets.sh cas-server-webapp/target/cas.war` reports 18 expected fixture results and 57 clean served files. Other checks disabled by the default nocheck profile are not claimed. The built WAR is retained as `../rev-cas5-r3.war`; all reviewer containers were removed; git diff check passes and the source checkout is clean. -Evidence: [build](../rev-cas5-r3-build.log), [fixture inputs](../rev-cas5-r3-mutants/cases.json), [scanner outcomes](../rev-cas5-r3-mutants.log), [browser harness](../rev-cas5-r3-browser.cjs), [browser outcomes](../rev-cas5-r3-browser.log), [extra cases and getLogo signature](../rev-cas5-r3-extra.log). Browser fixtures use installed headless Chrome, fulfill only their synthetic same-origin document, and abort every resource request. The author's S5 2-pass and fakecas 22-pass reruns were supplied in the task; they were not independently rerun or audited as new run folders here. No real server-rendered JSP, end-to-end login or deployed CSP was exercised in r3. +Evidence: [build](../../../rev-cas5-r3-build.log), [fixture inputs](../../../rev-cas5-r3-mutants/cases.json), [scanner outcomes](../../../rev-cas5-r3-mutants.log), [browser harness](../../../rev-cas5-r3-browser.cjs), [browser outcomes](../../../rev-cas5-r3-browser.log), [extra cases and getLogo signature](../../../rev-cas5-r3-extra.log). Browser fixtures use installed headless Chrome, fulfill only their synthetic same-origin document, and abort every resource request. The author's S5 2-pass and fakecas 22-pass reruns were supplied in the task; they were not independently rerun or audited as new run folders here. No real server-rendered JSP, end-to-end login or deployed CSP was exercised in r3. --- @@ -97,4 +97,4 @@ Evidence: [build](../rev-cas5-r3-build.log), [fixture inputs](../rev-cas5-r3-mut Dockerfile 固定の Maven/Temurin 8 digest の隔離コンテナに checkout を転送し、`mvn -B package -DskipTests=false` が1分31秒で成功。OAuth 109件・OSF 45件、失敗・エラー・skip なし。`sh etc/check-war-assets.sh cas-server-webapp/target/cas.war` は18 fixture が期待どおり、57配布ファイルは所見なしと報告した。既定 nocheck profile が無効にする他の検査の合格は主張しない。WAR は `../rev-cas5-r3.war` に保存し、レビュワのコンテナは全削除。git diff check 合格、ソース checkout は clean。 -証拠: [build](../rev-cas5-r3-build.log)、[入力](../rev-cas5-r3-mutants/cases.json)、[検査結果](../rev-cas5-r3-mutants.log)、[browser harness](../rev-cas5-r3-browser.cjs)、[browser 結果](../rev-cas5-r3-browser.log)、[追加試験と getLogo 型](../rev-cas5-r3-extra.log)。headless Chrome は模擬同一 origin 文書だけを返し、資産要求をすべて中止した。作者の S5 2合格・fakecas 22合格はタスクで提示された結果であり、今回独立再実行や新規実行フォルダーの監査はしていない。r3 では実 JSP 描画・ログイン全体・稼働 CSP を試験していない。 +証拠: [build](../../../rev-cas5-r3-build.log)、[入力](../../../rev-cas5-r3-mutants/cases.json)、[検査結果](../../../rev-cas5-r3-mutants.log)、[browser harness](../../../rev-cas5-r3-browser.cjs)、[browser 結果](../../../rev-cas5-r3-browser.log)、[追加試験と getLogo 型](../../../rev-cas5-r3-extra.log)。headless Chrome は模擬同一 origin 文書だけを返し、資産要求をすべて中止した。作者の S5 2合格・fakecas 22合格はタスクで提示された結果であり、今回独立再実行や新規実行フォルダーの監査はしていない。r3 では実 JSP 描画・ログイン全体・稼働 CSP を試験していない。 diff --git a/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r4.md b/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r4.md index e68d5e9d..0988047d 100644 --- a/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r4.md +++ b/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r4.md @@ -9,7 +9,7 @@ - Applies to: RDM-cas-overlay-dev2027 PR #5 `34a5ed6242f2233f3ea01ee70d79bb2749e6c418`, delta from `878099b0405060a0d05c0d52cb286e53abf381e1`. - Posted comment / 投稿コメント: issuecomment-6902722689. - Reviewer: REV(担当レビュワ, Codex Astra). -- Related: [Review index](../rev-guide88/docs/reviews/README.md) · [SPEC-06](../rev-guide88/docs/spec/06_documentation_conventions.md) · [r3](2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r3.md). Used by: CAS integration and guide PR #88 pin update. +- Related: [Review index](README.md) · [SPEC-06](../spec/06_documentation_conventions.md) · [r3](2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r3.md). Used by: CAS integration and guide PR #88 pin update. --- @@ -28,7 +28,7 @@ I recompiled the Java scanner in the Dockerfile's exact Maven/Temurin 8 image an Ordinary, non-obfuscated additions also fail as intended: external script src; link rel=stylesheet href; img src; CSS background url(); quoted and url()-form @import; @font-face src; a style block; inline style; and picture/source srcset with multiple candidates. No plain accidental external asset false negative was found in this matrix. The supplied fixtures additionally cover named entities, percent encoding and contextPath/backslash. The artifact guard checks duplicate occurrences instead of overwriting them, validates the contextPath suffix, and retains failure on scanner errors. -Evidence: [independent inputs](../rev-cas5-r4-mutants/cases.json) and [scanner results](../rev-cas5-r4-mutants.log). This is finite regression evidence, not a claim to recognize all possible HTML/CSS/JavaScript semantics. Further adversarial encodings outside the stated normalizations were deliberately not pursued under the revised assignment. +Evidence: [independent inputs](../../../rev-cas5-r4-mutants/cases.json) and [scanner results](../../../rev-cas5-r4-mutants.log). This is finite regression evidence, not a claim to recognize all possible HTML/CSS/JavaScript semantics. Further adversarial encodings outside the stated normalizations were deliberately not pursued under the revised assignment. ### Documentation and compatibility @@ -46,7 +46,7 @@ SHA-256 comparison against my retained r3 WAR found all 57 served-source files i No new browser/JSP or live S5/full-suite run was performed. Reusing the previous live results is reasonable for unchanged served assets, while retaining their prior coverage limits. The independently built artifact comparison does not certify arbitrary runtime configuration, every possible route, or overall deterministic suite behavior. Guide #88 still requires its separate dependency-pin integration; this review does not perform it. -Logs: [Maven and shipped guard](../rev-cas5-r4-build.log), [artifact comparison](../rev-cas5-r4-artifact-compare.json). The WAR is retained at `../rev-cas5-r4.war`. The source checkout remains unmodified; `git diff --check 878099b0..HEAD` passes, and reviewer containers were removed. +Logs: [Maven and shipped guard](../../../rev-cas5-r4-build.log), [artifact comparison](../../../rev-cas5-r4-artifact-compare.json). The WAR is retained at `../rev-cas5-r4.war`. The source checkout remains unmodified; `git diff --check 878099b0..HEAD` passes, and reviewer containers were removed. --- @@ -65,7 +65,7 @@ Dockerfile 固定の Maven/Temurin 8 イメージで Java 検査器を再コン 難読化していない通常の追加でも、script src、link rel=stylesheet href、img src、CSS background url()、引用符形式と url() 形式の @import、@font-face src、style 要素、inline style、picture/source の複数候補 srcset を正しく拒否した。この試験集合では通常の誤追加を見逃す例はなかった。同梱 fixture は名前付き文字参照・パーセント符号・contextPath/backslash も検査する。重複属性を上書きせず全件調べ、contextPath の後続を検証し、検査エラー時も失敗させる。 -証拠は [独立入力](../rev-cas5-r4-mutants/cases.json) と [結果](../rev-cas5-r4-mutants.log)。有限個の回帰試験であり、HTML/CSS/JavaScript の全構文を理解するという主張ではない。今回の指示に従い、明示された正規化の外にある追加の攻撃的符号化は追求していない。 +証拠は [独立入力](../../../rev-cas5-r4-mutants/cases.json) と [結果](../../../rev-cas5-r4-mutants.log)。有限個の回帰試験であり、HTML/CSS/JavaScript の全構文を理解するという主張ではない。今回の指示に従い、明示された正規化の外にある追加の攻撃的符号化は追求していない。 ### 文書と互換性 @@ -83,4 +83,4 @@ Dockerfile 固定イメージで `mvn -B package -DskipTests=false` が1分35秒 新たな browser/JSP 描画・実機 S5/full suite は実行していない。配布資産が変わらない範囲で前回実機結果を利用することは妥当だが、以前の検証範囲の限界は維持する。成果物比較は任意の実行時設定・全経路・スイート全体の決定性を証明しない。ガイド #88 の依存先固定版の統合は別途必要であり、このレビューでは行っていない。 -ログは [Maven と出荷検査](../rev-cas5-r4-build.log)、[成果物比較](../rev-cas5-r4-artifact-compare.json)。WAR は `../rev-cas5-r4.war` に保存。ソース checkout は未変更、`git diff --check 878099b0..HEAD` は合格、レビュワのコンテナは削除した。 +ログは [Maven と出荷検査](../../../rev-cas5-r4-build.log)、[成果物比較](../../../rev-cas5-r4-artifact-compare.json)。WAR は `../rev-cas5-r4.war` に保存。ソース checkout は未変更、`git diff --check 878099b0..HEAD` は合格、レビュワのコンテナは削除した。 diff --git a/docs/reviews/2026-11-16_RDM-developer-guide_PR85_Health_Settle_r1.md b/docs/reviews/2026-11-16_RDM-developer-guide_PR85_Health_Settle_r1.md index b8b579fb..ee808f60 100644 --- a/docs/reviews/2026-11-16_RDM-developer-guide_PR85_Health_Settle_r1.md +++ b/docs/reviews/2026-11-16_RDM-developer-guide_PR85_Health_Settle_r1.md @@ -7,7 +7,7 @@ - Version / 版: 0.1 — 2026-11-16. **Learned:** unknown uptime and proven long uptime need distinct outcomes. / 不明な稼働時間と、十分長いと確認できた稼働時間は区別する必要がある。 - Applies to: RCOSDP/RDM-developer-guide PR #85, head `e5bf62263dc0595e7acd4179a8f61a706249af77`, base `6b7cadb625e44f93fd5d7ce5d251f4aceada284d`. -- Related: [PR #85](https://github.com/RCOSDP/RDM-developer-guide/pull/85) · [Review index](../rev-pr85/docs/reviews/README.md) · [SPEC-02](../rev-pr85/docs/spec/02_installer_cli.md) · [SPEC-06](../rev-pr85/docs/spec/06_documentation_conventions.md) +- Related: [PR #85](https://github.com/RCOSDP/RDM-developer-guide/pull/85) · [Review index](README.md) · [SPEC-02](../spec/02_installer_cli.md) · [SPEC-06](../spec/06_documentation_conventions.md) --- diff --git a/docs/reviews/2026-11-16_RDM-developer-guide_PR85_Health_Settle_r2.md b/docs/reviews/2026-11-16_RDM-developer-guide_PR85_Health_Settle_r2.md index 5c15e024..6031b8c9 100644 --- a/docs/reviews/2026-11-16_RDM-developer-guide_PR85_Health_Settle_r2.md +++ b/docs/reviews/2026-11-16_RDM-developer-guide_PR85_Health_Settle_r2.md @@ -7,7 +7,7 @@ - Version / 版: 0.1 — 2026-11-16. **Learned:** explicit unknown-uptime rejection preserves the same readiness contract across polling and one-shot callers. / 稼働時間不明を明示的に拒否すれば、繰り返す呼び出し元と一回だけの呼び出し元で同じ健康判定の契約を保てる。 - Applies to: PR #85, head `5a0733c252a86a2dd2f139b8d75b7cd045c2bcf4`, base `6b7cadb625e44f93fd5d7ce5d251f4aceada284d`. -- Related: [PR #85](https://github.com/RCOSDP/RDM-developer-guide/pull/85) · [r1](2026-11-16_RDM-developer-guide_PR85_Health_Settle_r1.md) · [Review index](../rev-pr85/docs/reviews/README.md) · [SPEC-02](../rev-pr85/docs/spec/02_installer_cli.md) · [SPEC-06](../rev-pr85/docs/spec/06_documentation_conventions.md) +- Related: [PR #85](https://github.com/RCOSDP/RDM-developer-guide/pull/85) · [r1](2026-11-16_RDM-developer-guide_PR85_Health_Settle_r1.md) · [Review index](README.md) · [SPEC-02](../spec/02_installer_cli.md) · [SPEC-06](../spec/06_documentation_conventions.md) --- diff --git a/docs/reviews/2026-11-16_RDM-developer-guide_PR86_S3_aipf_OIDC_r1.md b/docs/reviews/2026-11-16_RDM-developer-guide_PR86_S3_aipf_OIDC_r1.md index a51ea419..ebeff2ef 100644 --- a/docs/reviews/2026-11-16_RDM-developer-guide_PR86_S3_aipf_OIDC_r1.md +++ b/docs/reviews/2026-11-16_RDM-developer-guide_PR86_S3_aipf_OIDC_r1.md @@ -7,7 +7,7 @@ - Version / 版: 0.1 — 2026-11-16. **Learned:** a test-body profile guard cannot prevent a module fixture from failing first. / 試験本体のプロファイル判定では、それより先に動くモジュール fixture の失敗を防げない。 - Applies to: RCOSDP/RDM-developer-guide PR #86, head `853053ede6b6d9d2c0550c2e3c3aa1056c53d370`, base `develop`; REV round 1. -- Related: [Review index](../rev-pr86/docs/reviews/README.md) · [SPEC-06](../rev-pr86/docs/spec/06_documentation_conventions.md) · [SPEC-04](../rev-pr86/docs/spec/04_acceptance_tests.md) · [ADR-15](../rev-pr86/docs/decisions/ADR-15_Mock_OIDC_Issuer_for_aipf.md). +- Related: [Review index](README.md) · [SPEC-06](../spec/06_documentation_conventions.md) · [SPEC-04](../spec/04_acceptance_tests.md) · [ADR-15](../decisions/ADR-15_Mock_OIDC_Issuer_for_aipf.md). - Used by: PR #86 revision and coordinator review decision. --- diff --git a/docs/reviews/2026-11-16_RDM-developer-guide_PR86_S3_aipf_OIDC_r2.md b/docs/reviews/2026-11-16_RDM-developer-guide_PR86_S3_aipf_OIDC_r2.md index 52d4de1a..562e46b5 100644 --- a/docs/reviews/2026-11-16_RDM-developer-guide_PR86_S3_aipf_OIDC_r2.md +++ b/docs/reviews/2026-11-16_RDM-developer-guide_PR86_S3_aipf_OIDC_r2.md @@ -7,7 +7,7 @@ - Version / 版: 0.1 — 2026-11-16. **Learned:** a fixture dependency enforces the eligibility check before key loading, and derived public settings remove the issuer's need to receive a signing key. / fixture の依存で鍵の読込みより先に対象判定を保証でき、公開設定の導出で発行者へ署名鍵を渡す必要も無くなる。 - Applies to: RCOSDP/RDM-developer-guide PR #86, head `6fccbd1`, base develop; REV round 2. -- Related: [Review index](../rev-pr86-r2/docs/reviews/README.md) · [SPEC-06](../rev-pr86-r2/docs/spec/06_documentation_conventions.md) · [ADR-15](../rev-pr86-r2/docs/decisions/ADR-15_Mock_OIDC_Issuer_for_aipf.md) · [r1](2026-11-16_RDM-developer-guide_PR86_S3_aipf_OIDC_r1.md) · [Upstream log review](2026-11-16_aipf_PR1_AccessLog_r1.md). +- Related: [Review index](README.md) · [SPEC-06](../spec/06_documentation_conventions.md) · [ADR-15](../decisions/ADR-15_Mock_OIDC_Issuer_for_aipf.md) · [r1](2026-11-16_RDM-developer-guide_PR86_S3_aipf_OIDC_r1.md) · [Upstream log review](2026-11-16_aipf_PR1_AccessLog_r1.md). - Used by: PR #86 integration decision. --- diff --git a/docs/reviews/2026-11-16_RDM-developer-guide_PR87_S2_rdmcp_r1.md b/docs/reviews/2026-11-16_RDM-developer-guide_PR87_S2_rdmcp_r1.md index 06eb1b5a..21be175e 100644 --- a/docs/reviews/2026-11-16_RDM-developer-guide_PR87_S2_rdmcp_r1.md +++ b/docs/reviews/2026-11-16_RDM-developer-guide_PR87_S2_rdmcp_r1.md @@ -8,7 +8,7 @@ - Version / 版: 0.1 — 2026-11-16. **Learned:** CAS uses whole-string service matching, and a real CAS bearer validator does not replace fakecas browser authentication. / CAS のサービス照合は文字列全体の一致であり、本物の CAS による Bearer 検証は fakecas のブラウザ認証を置き換えない。 - Applies to: RCOSDP/RDM-rdmcp-dev2027 PR #1 `9c6e0050468451bba0c0cb9e3f8fdbb45f0b310b` (A); RCOSDP/RDM-developer-guide PR #87 `e05f6205c5de382e9a00f3cf0945f4eac58a05bd` (B). - Reviewer: REV(担当レビュワ, Codex Astra) -- Related: [Review index](../rev-pr87/docs/reviews/README.md), [SPEC-06](../rev-pr87/docs/spec/06_documentation_conventions.md), [SPEC-04](../rev-pr87/docs/spec/04_acceptance_tests.md), [SPEC-02](../rev-pr87/docs/spec/02_installer_cli.md). +- Related: [Review index](README.md), [SPEC-06](../spec/06_documentation_conventions.md), [SPEC-04](../spec/04_acceptance_tests.md), [SPEC-02](../spec/02_installer_cli.md). - Used by: coordinator's round-1 merge decision; this record is uncommitted. --- @@ -58,7 +58,7 @@ checks[8]{check,result}: same Linux test on develop 7eed0db,"same failure: text/plain instead of presentation MIME" ``` -The failing test and MIME implementation are unchanged by A. The Linux baseline comparison confirms the reported pre-existing failure; macOS has the needed MIME mapping and passes. `tools/scan_secrets.py --all` in A found no actual values. Full Linux log: [rdmcp-linux-pytest.log](rdmcp-linux-pytest.log). Other logs: [installer](installer-tests.log), [tools](tools-tests.log), [docs](docs-check.log), [manifest](manifest-check.log), [macOS pytest](rdmcp-pytest.log). +The failing test and MIME implementation are unchanged by A. The Linux baseline comparison confirms the reported pre-existing failure; macOS has the needed MIME mapping and passes. `tools/scan_secrets.py --all` in A found no actual values. Full Linux log: `rdmcp-linux-pytest.log` (external evidence not kept in the repository). Other logs: installer (`installer-tests.log`, external evidence not kept in the repository), tools (`tools-tests.log`, external evidence not kept in the repository), docs (`docs-check.log`, external evidence not kept in the repository), manifest (`manifest-check.log`, external evidence not kept in the repository), macOS pytest (`rdmcp-pytest.log`, external evidence not kept in the repository). Both `gh pr view ... --json statusCheckRollup` and `gh run list --commit ` returned empty arrays for A and B. No completed or running CI was visible at review time; local checks must not be described as green CI. Both PRs were read using `gh pr diff`. No long polling was used. @@ -109,6 +109,6 @@ checks[8]{check,result}: develop 7eed0db の同じ Linux テスト,"同じ失敗: presentation MIME の代わりに text/plain" ``` -失敗したテストと MIME 実装は A で変更されていない。Linux の develop 比較により既存失敗との説明を確認した。macOS は必要な MIME 対応を持ち合格する。A の `tools/scan_secrets.py --all` は実値を検出しなかった。ログは [Linux pytest](rdmcp-linux-pytest.log)、[installer](installer-tests.log)、[tools](tools-tests.log)、[docs](docs-check.log)、[manifest](manifest-check.log)、[macOS pytest](rdmcp-pytest.log)。 +失敗したテストと MIME 実装は A で変更されていない。Linux の develop 比較により既存失敗との説明を確認した。macOS は必要な MIME 対応を持ち合格する。A の `tools/scan_secrets.py --all` は実値を検出しなかった。ログは Linux pytest (`rdmcp-linux-pytest.log`, external evidence not kept in the repository)、installer (`installer-tests.log`, external evidence not kept in the repository)、tools (`tools-tests.log`, external evidence not kept in the repository)、docs (`docs-check.log`, external evidence not kept in the repository)、manifest (`manifest-check.log`, external evidence not kept in the repository)、macOS pytest (`rdmcp-pytest.log`, external evidence not kept in the repository)。 A・B とも `gh pr view ... --json statusCheckRollup` と `gh run list --commit <対象SHA>` は空配列だった。確認時に完了・実行中 CI は表示されず、ローカル検査を CI 合格とは呼べない。両PRを `gh pr diff` で読み、長時間のポーリングはしていない。 diff --git a/docs/reviews/2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r1.md b/docs/reviews/2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r1.md index 00a2430d..465456cd 100644 --- a/docs/reviews/2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r1.md +++ b/docs/reviews/2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r1.md @@ -8,7 +8,7 @@ - Version / 版: 0.1 — 2026-11-16. **Learned:** a private Docker network prevents external exposure but does not prove that an identity header came from the SP. / Docker の内部ネットワークは外部公開を防ぐが、認証ヘッダが SP から来た証明にはならないと分かった。 - Applies to: RCOSDP/RDM-developer-guide PR #88, head `38c64143abd1ce367276ef5f53b52b3685457676`, base `develop`. - Reviewer: REV(担当レビュワ, Codex Astra). -- Related: [Review index](../rev-pr88/docs/reviews/README.md) · [SPEC-06](../rev-pr88/docs/spec/06_documentation_conventions.md) · [SPEC-04](../rev-pr88/docs/spec/04_acceptance_tests.md) · [ADR-05](../rev-pr88/docs/decisions/ADR-05_Test_IdP_for_Shibboleth.md). Used by: the coordinator's revision and merge decision. +- Related: [Review index](README.md) · [SPEC-06](../spec/06_documentation_conventions.md) · [SPEC-04](../spec/04_acceptance_tests.md) · [ADR-05](../decisions/ADR-05_Test_IdP_for_Shibboleth.md). Used by: the coordinator's revision and merge decision. --- diff --git a/docs/reviews/2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r2.md b/docs/reviews/2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r2.md index c65a235b..5efd75bd 100644 --- a/docs/reviews/2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r2.md +++ b/docs/reviews/2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r2.md @@ -8,7 +8,7 @@ - Version / 版: 0.1 — 2026-11-16. **Learned:** a tested upstream fix protects an installation only after its pinned dependency selects that fix. / 上流の修正が試験を通っても、固定した依存先がその修正を選ぶまでは配備を保護しない。 - Applies to: RDM-cas-overlay-dev2027 PR #4 `d9ad5c6661a3796aea63c9da487a9b51bb67b148` (A, r1), and RDM-developer-guide PR #88 `86b7cc752f78326f3edbdfa3b7a2745fe89bb747` (B, r2), both targeting develop. - Reviewer: REV(担当レビュワ, Codex Astra). -- Related: [Review index](../rev-pr88-r2/docs/reviews/README.md) · [SPEC-06](../rev-pr88-r2/docs/spec/06_documentation_conventions.md) · [SPEC-04](../rev-pr88-r2/docs/spec/04_acceptance_tests.md) · [r1](2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r1.md). Used by: the coordinator's revision and merge decision. +- Related: [Review index](README.md) · [SPEC-06](../spec/06_documentation_conventions.md) · [SPEC-04](../spec/04_acceptance_tests.md) · [r1](2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r1.md). Used by: the coordinator's revision and merge decision. --- diff --git a/docs/reviews/2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r3.md b/docs/reviews/2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r3.md index 97625ac2..02878df4 100644 --- a/docs/reviews/2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r3.md +++ b/docs/reviews/2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r3.md @@ -9,7 +9,7 @@ - Applies to: RDM-developer-guide PR #88 `25cccb29cc5c109afeebe43301d1f30bcaae701b`, delta from r2 `86b7cc752f78326f3edbdfa3b7a2745fe89bb747`. - Posted comment / 投稿コメント: issuecomment-6901736298. - Reviewer: REV(担当レビュワ, Codex Astra). -- Related: [Review index](../rev-guide88/docs/reviews/README.md) · [SPEC-06](../rev-guide88/docs/spec/06_documentation_conventions.md) · [r2](2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r2.md) · [CAS r2](2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r2.md). Used by: dependency integration and merge decision. +- Related: [Review index](README.md) · [SPEC-06](../spec/06_documentation_conventions.md) · [r2](2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r2.md) · [CAS r2](2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r2.md). Used by: dependency integration and merge decision. --- @@ -36,7 +36,7 @@ All ten newly committed files were read/scanned. No personal hostname, /Users or ### Checks and scope limits -From the detached fresh guide clone, `PYTHONPATH=installer python3 -m unittest installer.tests.test_accept` passed 71 tests (10 skipped); docs_check passed 223 files (8 exempt, 71 warnings); manifest_check and `git diff --check 86b7cc7..HEAD` passed. Logs: [unit](../rev-guide88-r3-unit.log), [docs](../rev-guide88-r3-docs.log), [manifest](../rev-guide88-r3-manifest.log). No need to rerun unchanged SP/CAS security wiring; the previous r2 assessment remains applicable, conditioned on the pin. No live stack, secrets, containers of the grdm-dev stack, CI polling, merge, approval, push or commit was performed. CAS's independent build is documented in its separate r2 record. +From the detached fresh guide clone, `PYTHONPATH=installer python3 -m unittest installer.tests.test_accept` passed 71 tests (10 skipped); docs_check passed 223 files (8 exempt, 71 warnings); manifest_check and `git diff --check 86b7cc7..HEAD` passed. Logs: [unit](../../../rev-guide88-r3-unit.log), [docs](../../../rev-guide88-r3-docs.log), [manifest](../../../rev-guide88-r3-manifest.log). No need to rerun unchanged SP/CAS security wiring; the previous r2 assessment remains applicable, conditioned on the pin. No live stack, secrets, containers of the grdm-dev stack, CI polling, merge, approval, push or commit was performed. CAS's independent build is documented in its separate r2 record. --- @@ -63,4 +63,4 @@ acc-20261115T211807Z-6bcf0e は results・collector・JUnit・pytest・再生成 ### 検査と範囲の限界 -fresh clone の detached head で `PYTHONPATH=installer python3 -m unittest installer.tests.test_accept` は71件、skip 10で合格。docs_check は223ファイル、除外8、警告71で合格。manifest_check と `git diff --check 86b7cc7..HEAD` も合格。ログは [unit](../rev-guide88-r3-unit.log)・[docs](../rev-guide88-r3-docs.log)・[manifest](../rev-guide88-r3-manifest.log)。変更のない SP/CAS 信頼配線は再試験しておらず、前回 r2 の評価を固定版条件付きで維持する。実機スタック・秘密・grdm-dev コンテナ操作、CI の継続ポーリング、merge・approve・push・commit はしていない。CAS の独立ビルドは別の r2 記録に記載した。 +fresh clone の detached head で `PYTHONPATH=installer python3 -m unittest installer.tests.test_accept` は71件、skip 10で合格。docs_check は223ファイル、除外8、警告71で合格。manifest_check と `git diff --check 86b7cc7..HEAD` も合格。ログは [unit](../../../rev-guide88-r3-unit.log)・[docs](../../../rev-guide88-r3-docs.log)・[manifest](../../../rev-guide88-r3-manifest.log)。変更のない SP/CAS 信頼配線は再試験しておらず、前回 r2 の評価を固定版条件付きで維持する。実機スタック・秘密・grdm-dev コンテナ操作、CI の継続ポーリング、merge・approve・push・commit はしていない。CAS の独立ビルドは別の r2 記録に記載した。 diff --git a/docs/reviews/2026-11-16_RDM-developer-guide_PR89_aipf_Pin_r1.md b/docs/reviews/2026-11-16_RDM-developer-guide_PR89_aipf_Pin_r1.md index eb77d149..a05565cc 100644 --- a/docs/reviews/2026-11-16_RDM-developer-guide_PR89_aipf_Pin_r1.md +++ b/docs/reviews/2026-11-16_RDM-developer-guide_PR89_aipf_Pin_r1.md @@ -8,7 +8,7 @@ - Version / 版: 0.1 — 2026-11-16. **Learned:** acceptance results establish scenario outcomes but do not by themselves record the running image identities. / 受入結果はシナリオの成否を示すが、それだけでは実行したイメージの識別子を記録しない。 - Applies to / 対象: RCOSDP/RDM-developer-guide PR #89, head `9c299c8422e0a537908741bc1550d17feab74400`, base `c77d48977e3b1c0329f31a7631eb964fd5d18085`. - Reviewer / 担当: REV(担当レビュワ, Codex Astra). -- Related / 関連: [Review index](../rev-pr89/docs/reviews/README.md) · [SPEC-06](../rev-pr89/docs/spec/06_documentation_conventions.md) · [Status board](../rev-pr89/docs/status/README.md). +- Related / 関連: [Review index](README.md) · [SPEC-06](../spec/06_documentation_conventions.md) · [Status board](../status/README.md). - Used by / 利用先: [PR #89 review decision](https://github.com/RCOSDP/RDM-developer-guide/pull/89). --- diff --git a/docs/reviews/2026-11-16_RDM-developer-guide_PR91_osf_Pin_c5a05de_r1.md b/docs/reviews/2026-11-16_RDM-developer-guide_PR91_osf_Pin_c5a05de_r1.md index aeb9ea6a..42e78b18 100644 --- a/docs/reviews/2026-11-16_RDM-developer-guide_PR91_osf_Pin_c5a05de_r1.md +++ b/docs/reviews/2026-11-16_RDM-developer-guide_PR91_osf_Pin_c5a05de_r1.md @@ -8,8 +8,8 @@ - Version / 版: 0.1 — 2026-11-16. **Learned:** the acceptance run proves core compatibility, while READ-user authorization needs separate upstream code and test evidence. / core の受入は互換性を示すが、READ の権限判定には上流のコードと試験という別の根拠が要る。 - Applies to: RCOSDP/RDM-developer-guide PR #91, head `368793d0b4677ad2c31eac4f342c69611b03c59c`, base `79ba329f538377378cd9f855856f97fbb70db605`. - Reviewer: REV(担当レビュワ, Codex Astra); independent round 1. -- Related: [review index](../rev-pr91/docs/reviews/README.md) · [SPEC-06](../rev-pr91/docs/spec/06_documentation_conventions.md) · [PR #91](https://github.com/RCOSDP/RDM-developer-guide/pull/91). -- Used by: [DEMO-20](../rev-pr91/docs/demo/20_core_install_mac_studio.md) · [status board](../rev-pr91/docs/status/README.md). +- Related: [review index](README.md) · [SPEC-06](../spec/06_documentation_conventions.md) · [PR #91](https://github.com/RCOSDP/RDM-developer-guide/pull/91). +- Used by: [DEMO-20](../demo/20_core_install_mac_studio.md) · [status board](../status/README.md). --- diff --git a/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r1.md b/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r1.md index bd936a20..c9b79029 100644 --- a/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r1.md +++ b/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r1.md @@ -7,7 +7,7 @@ - Version / 版: 0.1 — 2026-11-16. **Learned:** log redaction must recognize the same parameter names as the request parser. / ログのマスクは要求の解析器と同じパラメータ名を認識しなければならない。 - Applies to: RCOSDP/RDM-aipf_mcp_data_grdm-dev2027 PR #1, `0f472a024192459adf68454730cfae0e07e3d8e8`, base develop. -- Related: [Review index](../rev-pr86-r2/docs/reviews/README.md) · [SPEC-06](../rev-pr86-r2/docs/spec/06_documentation_conventions.md) · [Source](../rev-aipf-pr1/src/aipf_mcp_data_grdm/server.py) · [Tests](../rev-aipf-pr1/tests/test_access_log.py). +- Related: [Review index](README.md) · [SPEC-06](../spec/06_documentation_conventions.md) · [Source](../../../rev-aipf-pr1/src/aipf_mcp_data_grdm/server.py) · [Tests](../../../rev-aipf-pr1/tests/test_access_log.py). - Used by: PR #1 revision and coordinator review. --- diff --git a/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r2.md b/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r2.md index b27eb104..498ad9bc 100644 --- a/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r2.md +++ b/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r2.md @@ -7,7 +7,7 @@ - Version / 版: 0.1 — 2026-11-16. **Learned:** rewriting a logging format string must preserve its argument contract or the logging error path can disclose the original arguments. / ログの書式文字列を書き換えるとき引数との対応を壊すと、ログのエラー処理から元の引数が漏れる。 - Applies to: RCOSDP/RDM-aipf_mcp_data_grdm-dev2027 PR #1, head `01820f0`, base develop; REV round 2. -- Related: [Review index](../rev-pr86-r2/docs/reviews/README.md) · [SPEC-06](../rev-pr86-r2/docs/spec/06_documentation_conventions.md) · [r1](2026-11-16_aipf_PR1_AccessLog_r1.md) · [Source](../rev-aipf-pr1/src/aipf_mcp_data_grdm/server.py). +- Related: [Review index](README.md) · [SPEC-06](../spec/06_documentation_conventions.md) · [r1](2026-11-16_aipf_PR1_AccessLog_r1.md) · [Source](../../../rev-aipf-pr1/src/aipf_mcp_data_grdm/server.py). - Used by: PR #1 revision and coordinator decision. --- diff --git a/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r3.md b/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r3.md index 2b5e1207..191198bb 100644 --- a/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r3.md +++ b/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r3.md @@ -7,7 +7,7 @@ - Version / 版: 0.1 — 2026-11-16. **Learned:** a sanitized rendered message with empty arguments preserves literal percent signs without a second formatting pass. / 秘匿済みの整形結果と空の引数なら、再整形せずに percent 記号をそのまま保てる。 - Applies to: RCOSDP/RDM-aipf_mcp_data_grdm-dev2027 PR #1, head `25bf430`, base develop; REV round 3. -- Related: [Review index](../rev-pr86-r2/docs/reviews/README.md) · [SPEC-06](../rev-pr86-r2/docs/spec/06_documentation_conventions.md) · [r2](2026-11-16_aipf_PR1_AccessLog_r2.md) · [Source](../rev-aipf-pr1/src/aipf_mcp_data_grdm/server.py). +- Related: [Review index](README.md) · [SPEC-06](../spec/06_documentation_conventions.md) · [r2](2026-11-16_aipf_PR1_AccessLog_r2.md) · [Source](../../../rev-aipf-pr1/src/aipf_mcp_data_grdm/server.py). - Used by: PR #1 integration decision. --- From 9b927e9289312b67dea9e07a6b84f41149647965 Mon Sep 17 00:00:00 2001 From: Yusuke-KOMIYAMA Date: Mon, 16 Nov 2026 13:34:56 +0900 Subject: [PATCH 3/3] =?UTF-8?q?docs(review):=20=E3=83=AA=E3=83=9D=E3=82=B8?= =?UTF-8?q?=E3=83=88=E3=83=AA=E3=81=AE=E5=A4=96=E3=82=92=E6=8C=87=E3=81=99?= =?UTF-8?q?=E3=83=AA=E3=83=B3=E3=82=AF=2052=20=E4=BB=B6=E3=81=A8=E3=80=81?= =?UTF-8?q?=E5=90=8D=E5=89=8D=E3=81=A0=E3=81=91=E3=81=AE=E8=A1=8C=E3=81=8D?= =?UTF-8?q?=E5=85=88=206=20=E4=BB=B6=E3=82=92=E7=9B=B4=E3=81=99=EF=BC=88PR?= =?UTF-8?q?=20#92=20=E3=83=AC=E3=83=93=E3=83=A5=E3=83=BC=20r2=EF=BC=89?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit scratchpad の中の作業ツリーでは、リポジトリの外(別の作業ツリーや scratchpad)を指すリンクも 手元のファイルに当たってしまい、docs_check が通っていた。リポジトリの外を指すリンク 52 件は、 リンクを外して、リポジトリに置いていない外部の証跡だと明示した。PR22 r1 の、名前だけになった Markdown の行き先 6 件は、リンクでない書き方にした。判定と指摘の中身は変えていない。 リポジトリの追跡ファイルだけを別のディレクトリに写した複製で、docs_check と manifest_check の 合格を確かめた。 Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_01BW1YQk71jLXY2FaSzMnnYJ --- ...1-14_RDM-osf.io-dev2027_PR22_Private_Security_CI_r1.md | 6 +++--- ...2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r1.md | 4 ++-- ...2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r2.md | 4 ++-- ...2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r3.md | 4 ++-- ...2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r4.md | 8 ++++---- ...026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r3.md | 4 ++-- docs/reviews/2026-11-16_aipf_PR1_AccessLog_r1.md | 2 +- docs/reviews/2026-11-16_aipf_PR1_AccessLog_r2.md | 2 +- docs/reviews/2026-11-16_aipf_PR1_AccessLog_r3.md | 2 +- 9 files changed, 18 insertions(+), 18 deletions(-) diff --git a/docs/reviews/2026-11-14_RDM-osf.io-dev2027_PR22_Private_Security_CI_r1.md b/docs/reviews/2026-11-14_RDM-osf.io-dev2027_PR22_Private_Security_CI_r1.md index 48cb6946..c04cbc3c 100644 --- a/docs/reviews/2026-11-14_RDM-osf.io-dev2027_PR22_Private_Security_CI_r1.md +++ b/docs/reviews/2026-11-14_RDM-osf.io-dev2027_PR22_Private_Security_CI_r1.md @@ -8,7 +8,7 @@ - Version / 版: 0.1 — 2026-11-14. **Learned:** an implemented test task provides no continuous coverage until the workflow selects it; enabling the existing private task collects three otherwise omitted files without changing public-mode jobs. / 実装済みの試験タスクもワークフローから選ばれなければ継続検査にならず、既存の非公開タスクを有効にすると公開モードのジョブを変えずに未収集3ファイルを検査できる。 - Applies to: RCOSDP/RDM-osf.io-dev2027 PR #22, branch `ci/private-security-matrix`, head `3aad270c72577b273b52152a07628b60f0db687c`, base `develop` at `90b10661c8dcf039c18fdd214aef98191f97497c`. - Reviewer: REV(担当レビュワ, Codex Astra). -- Related: [Review index](README.md) · [SPEC-06](https://github.com/RCOSDP/RDM-developer-guide/blob/f005512dd059ea2d9077c36d41326d0b9b5b13af/docs/spec/06_documentation_conventions.md) · [PR #22](RDM-osf.io-dev2027 PR #22) · [Issue #16](RDM-osf.io-dev2027 issue #16). +- Related: [Review index](README.md) · [SPEC-06](https://github.com/RCOSDP/RDM-developer-guide/blob/f005512dd059ea2d9077c36d41326d0b9b5b13af/docs/spec/06_documentation_conventions.md) · PR #22 (RDM-osf.io-dev2027 PR #22) · Issue #16 (RDM-osf.io-dev2027 issue #16). - Used by: issue #16 completion assessment and the coordinator's merge decision. Internal review record; private repository links are intended for authorized readers. --- @@ -51,7 +51,7 @@ private_security_results[2]{event,run,job,conclusion,passed,warnings,test_second push,43562240653,132356584468,success,351,76876,587.04,2026-11-14T06:28:38Z ``` -Evidence: [PR-event job](RDM-osf.io-dev2027 Actions run 43562245773), [push-event job](RDM-osf.io-dev2027 Actions run 43562240653), saved PR log (`PR22_r1_private_security_pr.log`, external evidence not kept in the repository), saved push log (`PR22_r1_private_security_push.log`, external evidence not kept in the repository). The warning counts are reported separately from failures. No failing tests or failure cause remain to diagnose. +Evidence: PR-event job (RDM-osf.io-dev2027 Actions run 43562245773), push-event job (RDM-osf.io-dev2027 Actions run 43562240653), saved PR log (`PR22_r1_private_security_pr.log`, external evidence not kept in the repository), saved push log (`PR22_r1_private_security_push.log`, external evidence not kept in the repository). The warning counts are reported separately from failures. No failing tests or failure cause remain to diagnose. ### Scope and remaining work @@ -97,7 +97,7 @@ private_security_results[2]{event,run,job,conclusion,passed,warnings,test_second push,43562240653,132356584468,success,351,76876,587.04,2026-11-14T06:28:38Z ``` -証拠: [PR 起点ジョブ](RDM-osf.io-dev2027 Actions run 43562245773)、[push 起点ジョブ](RDM-osf.io-dev2027 Actions run 43562240653)、保存した PR ログ (`PR22_r1_private_security_pr.log`, external evidence not kept in the repository)、保存した push ログ (`PR22_r1_private_security_push.log`, external evidence not kept in the repository)。警告件数は失敗件数と区別して記録した。原因調査が必要な失敗試験はない。 +証拠: PR 起点ジョブ (RDM-osf.io-dev2027 Actions run 43562245773)、push 起点ジョブ (RDM-osf.io-dev2027 Actions run 43562240653)、保存した PR ログ (`PR22_r1_private_security_pr.log`, external evidence not kept in the repository)、保存した push ログ (`PR22_r1_private_security_push.log`, external evidence not kept in the repository)。警告件数は失敗件数と区別して記録した。原因調査が必要な失敗試験はない。 ### 範囲と残作業 diff --git a/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r1.md b/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r1.md index c09d04b3..35fb6e7f 100644 --- a/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r1.md +++ b/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r1.md @@ -47,7 +47,7 @@ The independent mock-DOM probe passed DOMContentLoaded and already-ready initial ### Evidence and verification limits -Evidence files are outside the clone: [source scan](../../../rev-cas5-scan.log), [mock-DOM probe](../../../rev-cas5-probe.cjs), [probe results](../../../rev-cas5-probe.log), [headless browser probe](../../../rev-cas5-browser.cjs), [browser results](../../../rev-cas5-browser.log), and [Maven build log](../../../rev-cas5-build.log). The probes run from the scratchpad directory with `node rev-cas5-probe.cjs` and `node rev-cas5-browser.cjs`; the latter imports the locally installed Playwright package and launches installed Chrome. All browser network requests are intercepted: only the synthetic same-origin HTML and exact cas.js bytes are fulfilled, and every other request is aborted. Monitoring fixtures intentionally omit JSP rendering and reproduce the extracted JavaScript failures. +Evidence files are outside the clone: source scan (`../../../rev-cas5-scan.log`, external evidence not kept in the repository), mock-DOM probe (`../../../rev-cas5-probe.cjs`, external evidence not kept in the repository), probe results (`../../../rev-cas5-probe.log`, external evidence not kept in the repository), headless browser probe (`../../../rev-cas5-browser.cjs`, external evidence not kept in the repository), browser results (`../../../rev-cas5-browser.log`, external evidence not kept in the repository), and Maven build log (`../../../rev-cas5-build.log`, external evidence not kept in the repository). The probes run from the scratchpad directory with `node rev-cas5-probe.cjs` and `node rev-cas5-browser.cjs`; the latter imports the locally installed Playwright package and launches installed Chrome. All browser network requests are intercepted: only the synthetic same-origin HTML and exact cas.js bytes are fulfilled, and every other request is aborted. Monitoring fixtures intentionally omit JSP rendering and reproduce the extracted JavaScript failures. Local Maven was absent and the host Java launcher reported no runtime. An initial bind-mounted container could not see the local POM, so the fresh checkout was streamed with tar into an isolated, disposable Maven/Temurin 8 container using the Dockerfile-pinned digest `sha256:0f402712b7a469189cfea5ec69cdcadcb32b22aaf1ea62777ab925c3d8470175`. `mvn -B -DskipTests=false package` succeeded in 1:55, with 109 OAuth tests and 41 OSF tests passing (150 total, no failures/errors/skips), and generated `/code/cas-server-webapp/target/cas.war`. The default nocheck profile still disables other static checks; this is not a claim those checks ran. The disposable container was automatically removed; the WAR was not retained or deployed. No standalone CAS runtime with its backend configuration was provisioned, so servlet-generated output remains unverified. The downloaded upstream WAR has SHA-256 `78d1f99f9eeacffeb6df0bb475490539c7509e3698cce60894a76276c1d34030`. @@ -88,7 +88,7 @@ No real server-rendered login/institution/OTP/error JSP was loaded with network ### 証拠と未検証範囲 -証拠は clone 外に保存した。[参照検索](../../../rev-cas5-scan.log)、[模擬 DOM 試験](../../../rev-cas5-probe.cjs)、[試験結果](../../../rev-cas5-probe.log)、[headless browser 試験](../../../rev-cas5-browser.cjs)、[browser 結果](../../../rev-cas5-browser.log)、[Maven ビルドログ](../../../rev-cas5-build.log)。scratchpad ディレクトリで `node rev-cas5-probe.cjs` と `node rev-cas5-browser.cjs` を実行した。後者はローカルの Playwright とインストール済み Chrome を使用する。ブラウザ通信は全件捕捉し、模擬 HTML と cas.js の実ファイルだけを同一 origin として返し、それ以外を中止する。監視画面の fixture は JSP を描画せず、抽出した JavaScript の失敗を再現する。 +証拠は clone 外に保存した。参照検索 (`../../../rev-cas5-scan.log`, external evidence not kept in the repository)、模擬 DOM 試験 (`../../../rev-cas5-probe.cjs`, external evidence not kept in the repository)、試験結果 (`../../../rev-cas5-probe.log`, external evidence not kept in the repository)、headless browser 試験 (`../../../rev-cas5-browser.cjs`, external evidence not kept in the repository)、browser 結果 (`../../../rev-cas5-browser.log`, external evidence not kept in the repository)、Maven ビルドログ (`../../../rev-cas5-build.log`, external evidence not kept in the repository)。scratchpad ディレクトリで `node rev-cas5-probe.cjs` と `node rev-cas5-browser.cjs` を実行した。後者はローカルの Playwright とインストール済み Chrome を使用する。ブラウザ通信は全件捕捉し、模擬 HTML と cas.js の実ファイルだけを同一 origin として返し、それ以外を中止する。監視画面の fixture は JSP を描画せず、抽出した JavaScript の失敗を再現する。 ホストには Maven がなく、Java ランチャーも実行環境なしと報告した。最初の bind mount コンテナにはローカル POM が見えなかったため、fresh checkout を tar で隔離した使い捨て Maven/Temurin 8 コンテナへ転送した。イメージは Dockerfile が固定する digest `sha256:0f402712b7a469189cfea5ec69cdcadcb32b22aaf1ea62777ab925c3d8470175`。`mvn -B -DskipTests=false package` は1分55秒で成功し、OAuth 109件・OSF 41件の計150件が失敗・エラー・skip なしで合格、`/code/cas-server-webapp/target/cas.war` を生成した。既定の nocheck profile は他の静的検査を無効にするため、それらの実行成功は主張しない。使い捨てコンテナは自動削除され、生成 WAR は保存・配備していない。バックエンド設定を備えた独立 CAS 実行環境は用意しておらず、サーブレット生成結果は未検証である。取得した上流 WAR の SHA-256 は `78d1f99f9eeacffeb6df0bb475490539c7509e3698cce60894a76276c1d34030`。 diff --git a/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r2.md b/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r2.md index 5a7d0b79..c23484db 100644 --- a/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r2.md +++ b/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r2.md @@ -53,7 +53,7 @@ No `/status/config` or `/statistics/ssosessions` dependency was found in guide d The Dockerfile-pinned Maven/Temurin 8 digest was used in an isolated container. From a tar-streamed copy of the checkout, `mvn -B -DskipTests=false package` succeeded in 1:32, with OAuth 109 and OSF 45 tests (154 total, no failures/errors/skips), and the exact shipped check accepted the built WAR: 57 served files. Other checks disabled by the default nocheck profile are not claimed. The retained WAR is `../rev-cas5-r2.war`; its static extraction is `../rev-cas5-r2-served`. Reviewer containers were removed. -The unmodified upstream CAS 4.1.5 WAR was also rejected (10 unique lines). This is a different artifact from the author's old overlay WAR; the claimed count of eight for that artifact was not independently reproduced. Evidence: [build log](../../../rev-cas5-r2-build.log), [static scan](../../../rev-cas5-r2-served-scan.log), [mutation results](../../../rev-cas5-r2-mutants.log), [fixture inputs](../../../rev-cas5-r2-mutants/cases.json), [old artifact and space-filename results](../../../rev-cas5-r2-old-check.log). No live JSP rendering or stack acceptance was rerun; guide evidence is reviewed separately. +The unmodified upstream CAS 4.1.5 WAR was also rejected (10 unique lines). This is a different artifact from the author's old overlay WAR; the claimed count of eight for that artifact was not independently reproduced. Evidence: build log (`../../../rev-cas5-r2-build.log`, external evidence not kept in the repository), static scan (`../../../rev-cas5-r2-served-scan.log`, external evidence not kept in the repository), mutation results (`../../../rev-cas5-r2-mutants.log`, external evidence not kept in the repository), fixture inputs (`../../../rev-cas5-r2-mutants/cases.json`, external evidence not kept in the repository), old artifact and space-filename results (`../../../rev-cas5-r2-old-check.log`, external evidence not kept in the repository). No live JSP rendering or stack acceptance was rerun; guide evidence is reviewed separately. --- @@ -86,4 +86,4 @@ Google Fonts・html5shiv・Analytics の読み込みは除去され、CSS はシ Dockerfile 固定の Maven/Temurin 8 digest の隔離コンテナへ checkout を tar 転送し、`mvn -B -DskipTests=false package` が1分32秒で成功した。OAuth 109件・OSF 45件、合計154件が失敗・エラー・skip なし。出荷する検査スクリプトは生成 WAR の57ファイルを合格と判定した。既定 nocheck profile が無効化する他の検査の成功は主張しない。WAR は `../rev-cas5-r2.war`、展開結果は `../rev-cas5-r2-served` に保存し、レビュワのコンテナは削除した。 -未変更の上流 CAS 4.1.5 WAR も拒否した(重複除去後10行)。これは作者の旧 overlay WAR とは別であり、作者の8行という件数は独立再現していない。証拠は [build log](../../../rev-cas5-r2-build.log)、[静的検索](../../../rev-cas5-r2-served-scan.log)、[変異試験結果](../../../rev-cas5-r2-mutants.log)、[入力](../../../rev-cas5-r2-mutants/cases.json)、[旧成果物と空白ファイル名](../../../rev-cas5-r2-old-check.log)。実 JSP 描画とスタック受入は再実行せず、ガイドの証拠は別レビューで扱う。 +未変更の上流 CAS 4.1.5 WAR も拒否した(重複除去後10行)。これは作者の旧 overlay WAR とは別であり、作者の8行という件数は独立再現していない。証拠は build log (`../../../rev-cas5-r2-build.log`, external evidence not kept in the repository)、静的検索 (`../../../rev-cas5-r2-served-scan.log`, external evidence not kept in the repository)、変異試験結果 (`../../../rev-cas5-r2-mutants.log`, external evidence not kept in the repository)、入力 (`../../../rev-cas5-r2-mutants/cases.json`, external evidence not kept in the repository)、旧成果物と空白ファイル名 (`../../../rev-cas5-r2-old-check.log`, external evidence not kept in the repository)。実 JSP 描画とスタック受入は再実行せず、ガイドの証拠は別レビューで扱う。 diff --git a/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r3.md b/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r3.md index 3355773e..4c176432 100644 --- a/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r3.md +++ b/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r3.md @@ -58,7 +58,7 @@ Either explicitly document intentional removal of all service logos, or provide The checkout was streamed into an isolated container using the exact Dockerfile-pinned Maven/Temurin 8 digest. `mvn -B package -DskipTests=false` succeeded in 1:31, OAuth 109 and OSF 45 tests, with zero failures/errors/skips. `sh etc/check-war-assets.sh cas-server-webapp/target/cas.war` reports 18 expected fixture results and 57 clean served files. Other checks disabled by the default nocheck profile are not claimed. The built WAR is retained as `../rev-cas5-r3.war`; all reviewer containers were removed; git diff check passes and the source checkout is clean. -Evidence: [build](../../../rev-cas5-r3-build.log), [fixture inputs](../../../rev-cas5-r3-mutants/cases.json), [scanner outcomes](../../../rev-cas5-r3-mutants.log), [browser harness](../../../rev-cas5-r3-browser.cjs), [browser outcomes](../../../rev-cas5-r3-browser.log), [extra cases and getLogo signature](../../../rev-cas5-r3-extra.log). Browser fixtures use installed headless Chrome, fulfill only their synthetic same-origin document, and abort every resource request. The author's S5 2-pass and fakecas 22-pass reruns were supplied in the task; they were not independently rerun or audited as new run folders here. No real server-rendered JSP, end-to-end login or deployed CSP was exercised in r3. +Evidence: build (`../../../rev-cas5-r3-build.log`, external evidence not kept in the repository), fixture inputs (`../../../rev-cas5-r3-mutants/cases.json`, external evidence not kept in the repository), scanner outcomes (`../../../rev-cas5-r3-mutants.log`, external evidence not kept in the repository), browser harness (`../../../rev-cas5-r3-browser.cjs`, external evidence not kept in the repository), browser outcomes (`../../../rev-cas5-r3-browser.log`, external evidence not kept in the repository), extra cases and getLogo signature (`../../../rev-cas5-r3-extra.log`, external evidence not kept in the repository). Browser fixtures use installed headless Chrome, fulfill only their synthetic same-origin document, and abort every resource request. The author's S5 2-pass and fakecas 22-pass reruns were supplied in the task; they were not independently rerun or audited as new run folders here. No real server-rendered JSP, end-to-end login or deployed CSP was exercised in r3. --- @@ -97,4 +97,4 @@ Evidence: [build](../../../rev-cas5-r3-build.log), [fixture inputs](../../../rev Dockerfile 固定の Maven/Temurin 8 digest の隔離コンテナに checkout を転送し、`mvn -B package -DskipTests=false` が1分31秒で成功。OAuth 109件・OSF 45件、失敗・エラー・skip なし。`sh etc/check-war-assets.sh cas-server-webapp/target/cas.war` は18 fixture が期待どおり、57配布ファイルは所見なしと報告した。既定 nocheck profile が無効にする他の検査の合格は主張しない。WAR は `../rev-cas5-r3.war` に保存し、レビュワのコンテナは全削除。git diff check 合格、ソース checkout は clean。 -証拠: [build](../../../rev-cas5-r3-build.log)、[入力](../../../rev-cas5-r3-mutants/cases.json)、[検査結果](../../../rev-cas5-r3-mutants.log)、[browser harness](../../../rev-cas5-r3-browser.cjs)、[browser 結果](../../../rev-cas5-r3-browser.log)、[追加試験と getLogo 型](../../../rev-cas5-r3-extra.log)。headless Chrome は模擬同一 origin 文書だけを返し、資産要求をすべて中止した。作者の S5 2合格・fakecas 22合格はタスクで提示された結果であり、今回独立再実行や新規実行フォルダーの監査はしていない。r3 では実 JSP 描画・ログイン全体・稼働 CSP を試験していない。 +証拠: build (`../../../rev-cas5-r3-build.log`, external evidence not kept in the repository)、入力 (`../../../rev-cas5-r3-mutants/cases.json`, external evidence not kept in the repository)、検査結果 (`../../../rev-cas5-r3-mutants.log`, external evidence not kept in the repository)、browser harness (`../../../rev-cas5-r3-browser.cjs`, external evidence not kept in the repository)、browser 結果 (`../../../rev-cas5-r3-browser.log`, external evidence not kept in the repository)、追加試験と getLogo 型 (`../../../rev-cas5-r3-extra.log`, external evidence not kept in the repository)。headless Chrome は模擬同一 origin 文書だけを返し、資産要求をすべて中止した。作者の S5 2合格・fakecas 22合格はタスクで提示された結果であり、今回独立再実行や新規実行フォルダーの監査はしていない。r3 では実 JSP 描画・ログイン全体・稼働 CSP を試験していない。 diff --git a/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r4.md b/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r4.md index 0988047d..68c4796d 100644 --- a/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r4.md +++ b/docs/reviews/2026-11-16_RDM-cas-overlay_PR5_NoExternalScripts_r4.md @@ -28,7 +28,7 @@ I recompiled the Java scanner in the Dockerfile's exact Maven/Temurin 8 image an Ordinary, non-obfuscated additions also fail as intended: external script src; link rel=stylesheet href; img src; CSS background url(); quoted and url()-form @import; @font-face src; a style block; inline style; and picture/source srcset with multiple candidates. No plain accidental external asset false negative was found in this matrix. The supplied fixtures additionally cover named entities, percent encoding and contextPath/backslash. The artifact guard checks duplicate occurrences instead of overwriting them, validates the contextPath suffix, and retains failure on scanner errors. -Evidence: [independent inputs](../../../rev-cas5-r4-mutants/cases.json) and [scanner results](../../../rev-cas5-r4-mutants.log). This is finite regression evidence, not a claim to recognize all possible HTML/CSS/JavaScript semantics. Further adversarial encodings outside the stated normalizations were deliberately not pursued under the revised assignment. +Evidence: independent inputs (`../../../rev-cas5-r4-mutants/cases.json`, external evidence not kept in the repository) and scanner results (`../../../rev-cas5-r4-mutants.log`, external evidence not kept in the repository). This is finite regression evidence, not a claim to recognize all possible HTML/CSS/JavaScript semantics. Further adversarial encodings outside the stated normalizations were deliberately not pursued under the revised assignment. ### Documentation and compatibility @@ -46,7 +46,7 @@ SHA-256 comparison against my retained r3 WAR found all 57 served-source files i No new browser/JSP or live S5/full-suite run was performed. Reusing the previous live results is reasonable for unchanged served assets, while retaining their prior coverage limits. The independently built artifact comparison does not certify arbitrary runtime configuration, every possible route, or overall deterministic suite behavior. Guide #88 still requires its separate dependency-pin integration; this review does not perform it. -Logs: [Maven and shipped guard](../../../rev-cas5-r4-build.log), [artifact comparison](../../../rev-cas5-r4-artifact-compare.json). The WAR is retained at `../rev-cas5-r4.war`. The source checkout remains unmodified; `git diff --check 878099b0..HEAD` passes, and reviewer containers were removed. +Logs: Maven and shipped guard (`../../../rev-cas5-r4-build.log`, external evidence not kept in the repository), artifact comparison (`../../../rev-cas5-r4-artifact-compare.json`, external evidence not kept in the repository). The WAR is retained at `../rev-cas5-r4.war`. The source checkout remains unmodified; `git diff --check 878099b0..HEAD` passes, and reviewer containers were removed. --- @@ -65,7 +65,7 @@ Dockerfile 固定の Maven/Temurin 8 イメージで Java 検査器を再コン 難読化していない通常の追加でも、script src、link rel=stylesheet href、img src、CSS background url()、引用符形式と url() 形式の @import、@font-face src、style 要素、inline style、picture/source の複数候補 srcset を正しく拒否した。この試験集合では通常の誤追加を見逃す例はなかった。同梱 fixture は名前付き文字参照・パーセント符号・contextPath/backslash も検査する。重複属性を上書きせず全件調べ、contextPath の後続を検証し、検査エラー時も失敗させる。 -証拠は [独立入力](../../../rev-cas5-r4-mutants/cases.json) と [結果](../../../rev-cas5-r4-mutants.log)。有限個の回帰試験であり、HTML/CSS/JavaScript の全構文を理解するという主張ではない。今回の指示に従い、明示された正規化の外にある追加の攻撃的符号化は追求していない。 +証拠は 独立入力 (`../../../rev-cas5-r4-mutants/cases.json`, external evidence not kept in the repository) と 結果 (`../../../rev-cas5-r4-mutants.log`, external evidence not kept in the repository)。有限個の回帰試験であり、HTML/CSS/JavaScript の全構文を理解するという主張ではない。今回の指示に従い、明示された正規化の外にある追加の攻撃的符号化は追求していない。 ### 文書と互換性 @@ -83,4 +83,4 @@ Dockerfile 固定イメージで `mvn -B package -DskipTests=false` が1分35秒 新たな browser/JSP 描画・実機 S5/full suite は実行していない。配布資産が変わらない範囲で前回実機結果を利用することは妥当だが、以前の検証範囲の限界は維持する。成果物比較は任意の実行時設定・全経路・スイート全体の決定性を証明しない。ガイド #88 の依存先固定版の統合は別途必要であり、このレビューでは行っていない。 -ログは [Maven と出荷検査](../../../rev-cas5-r4-build.log)、[成果物比較](../../../rev-cas5-r4-artifact-compare.json)。WAR は `../rev-cas5-r4.war` に保存。ソース checkout は未変更、`git diff --check 878099b0..HEAD` は合格、レビュワのコンテナは削除した。 +ログは Maven と出荷検査 (`../../../rev-cas5-r4-build.log`, external evidence not kept in the repository)、成果物比較 (`../../../rev-cas5-r4-artifact-compare.json`, external evidence not kept in the repository)。WAR は `../rev-cas5-r4.war` に保存。ソース checkout は未変更、`git diff --check 878099b0..HEAD` は合格、レビュワのコンテナは削除した。 diff --git a/docs/reviews/2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r3.md b/docs/reviews/2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r3.md index 02878df4..ca84beb8 100644 --- a/docs/reviews/2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r3.md +++ b/docs/reviews/2026-11-16_RDM-developer-guide_PR88_S5_Shibboleth_r3.md @@ -36,7 +36,7 @@ All ten newly committed files were read/scanned. No personal hostname, /Users or ### Checks and scope limits -From the detached fresh guide clone, `PYTHONPATH=installer python3 -m unittest installer.tests.test_accept` passed 71 tests (10 skipped); docs_check passed 223 files (8 exempt, 71 warnings); manifest_check and `git diff --check 86b7cc7..HEAD` passed. Logs: [unit](../../../rev-guide88-r3-unit.log), [docs](../../../rev-guide88-r3-docs.log), [manifest](../../../rev-guide88-r3-manifest.log). No need to rerun unchanged SP/CAS security wiring; the previous r2 assessment remains applicable, conditioned on the pin. No live stack, secrets, containers of the grdm-dev stack, CI polling, merge, approval, push or commit was performed. CAS's independent build is documented in its separate r2 record. +From the detached fresh guide clone, `PYTHONPATH=installer python3 -m unittest installer.tests.test_accept` passed 71 tests (10 skipped); docs_check passed 223 files (8 exempt, 71 warnings); manifest_check and `git diff --check 86b7cc7..HEAD` passed. Logs: unit (`../../../rev-guide88-r3-unit.log`, external evidence not kept in the repository), docs (`../../../rev-guide88-r3-docs.log`, external evidence not kept in the repository), manifest (`../../../rev-guide88-r3-manifest.log`, external evidence not kept in the repository). No need to rerun unchanged SP/CAS security wiring; the previous r2 assessment remains applicable, conditioned on the pin. No live stack, secrets, containers of the grdm-dev stack, CI polling, merge, approval, push or commit was performed. CAS's independent build is documented in its separate r2 record. --- @@ -63,4 +63,4 @@ acc-20261115T211807Z-6bcf0e は results・collector・JUnit・pytest・再生成 ### 検査と範囲の限界 -fresh clone の detached head で `PYTHONPATH=installer python3 -m unittest installer.tests.test_accept` は71件、skip 10で合格。docs_check は223ファイル、除外8、警告71で合格。manifest_check と `git diff --check 86b7cc7..HEAD` も合格。ログは [unit](../../../rev-guide88-r3-unit.log)・[docs](../../../rev-guide88-r3-docs.log)・[manifest](../../../rev-guide88-r3-manifest.log)。変更のない SP/CAS 信頼配線は再試験しておらず、前回 r2 の評価を固定版条件付きで維持する。実機スタック・秘密・grdm-dev コンテナ操作、CI の継続ポーリング、merge・approve・push・commit はしていない。CAS の独立ビルドは別の r2 記録に記載した。 +fresh clone の detached head で `PYTHONPATH=installer python3 -m unittest installer.tests.test_accept` は71件、skip 10で合格。docs_check は223ファイル、除外8、警告71で合格。manifest_check と `git diff --check 86b7cc7..HEAD` も合格。ログは unit (`../../../rev-guide88-r3-unit.log`, external evidence not kept in the repository)・docs (`../../../rev-guide88-r3-docs.log`, external evidence not kept in the repository)・manifest (`../../../rev-guide88-r3-manifest.log`, external evidence not kept in the repository)。変更のない SP/CAS 信頼配線は再試験しておらず、前回 r2 の評価を固定版条件付きで維持する。実機スタック・秘密・grdm-dev コンテナ操作、CI の継続ポーリング、merge・approve・push・commit はしていない。CAS の独立ビルドは別の r2 記録に記載した。 diff --git a/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r1.md b/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r1.md index c9b79029..576c1338 100644 --- a/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r1.md +++ b/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r1.md @@ -7,7 +7,7 @@ - Version / 版: 0.1 — 2026-11-16. **Learned:** log redaction must recognize the same parameter names as the request parser. / ログのマスクは要求の解析器と同じパラメータ名を認識しなければならない。 - Applies to: RCOSDP/RDM-aipf_mcp_data_grdm-dev2027 PR #1, `0f472a024192459adf68454730cfae0e07e3d8e8`, base develop. -- Related: [Review index](README.md) · [SPEC-06](../spec/06_documentation_conventions.md) · [Source](../../../rev-aipf-pr1/src/aipf_mcp_data_grdm/server.py) · [Tests](../../../rev-aipf-pr1/tests/test_access_log.py). +- Related: [Review index](README.md) · [SPEC-06](../spec/06_documentation_conventions.md) · Source (`../../../rev-aipf-pr1/src/aipf_mcp_data_grdm/server.py`, external evidence not kept in the repository) · Tests (`../../../rev-aipf-pr1/tests/test_access_log.py`, external evidence not kept in the repository). - Used by: PR #1 revision and coordinator review. --- diff --git a/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r2.md b/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r2.md index 498ad9bc..6ef83c35 100644 --- a/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r2.md +++ b/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r2.md @@ -7,7 +7,7 @@ - Version / 版: 0.1 — 2026-11-16. **Learned:** rewriting a logging format string must preserve its argument contract or the logging error path can disclose the original arguments. / ログの書式文字列を書き換えるとき引数との対応を壊すと、ログのエラー処理から元の引数が漏れる。 - Applies to: RCOSDP/RDM-aipf_mcp_data_grdm-dev2027 PR #1, head `01820f0`, base develop; REV round 2. -- Related: [Review index](README.md) · [SPEC-06](../spec/06_documentation_conventions.md) · [r1](2026-11-16_aipf_PR1_AccessLog_r1.md) · [Source](../../../rev-aipf-pr1/src/aipf_mcp_data_grdm/server.py). +- Related: [Review index](README.md) · [SPEC-06](../spec/06_documentation_conventions.md) · [r1](2026-11-16_aipf_PR1_AccessLog_r1.md) · Source (`../../../rev-aipf-pr1/src/aipf_mcp_data_grdm/server.py`, external evidence not kept in the repository). - Used by: PR #1 revision and coordinator decision. --- diff --git a/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r3.md b/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r3.md index 191198bb..d4526eba 100644 --- a/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r3.md +++ b/docs/reviews/2026-11-16_aipf_PR1_AccessLog_r3.md @@ -7,7 +7,7 @@ - Version / 版: 0.1 — 2026-11-16. **Learned:** a sanitized rendered message with empty arguments preserves literal percent signs without a second formatting pass. / 秘匿済みの整形結果と空の引数なら、再整形せずに percent 記号をそのまま保てる。 - Applies to: RCOSDP/RDM-aipf_mcp_data_grdm-dev2027 PR #1, head `25bf430`, base develop; REV round 3. -- Related: [Review index](README.md) · [SPEC-06](../spec/06_documentation_conventions.md) · [r2](2026-11-16_aipf_PR1_AccessLog_r2.md) · [Source](../../../rev-aipf-pr1/src/aipf_mcp_data_grdm/server.py). +- Related: [Review index](README.md) · [SPEC-06](../spec/06_documentation_conventions.md) · [r2](2026-11-16_aipf_PR1_AccessLog_r2.md) · Source (`../../../rev-aipf-pr1/src/aipf_mcp_data_grdm/server.py`, external evidence not kept in the repository). - Used by: PR #1 integration decision. ---