Repository navigation
Validate native Requirements bootstrap context and tool files - #6
Conversation
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
|
Independent shipping verification returned PASS+NOTES for head Root read the complete frozen report and accepted its narrow foundation verdict. Report SHA256 is The owner authorized a normal merge that preserves this immutable policy revision as an ancestor. This landing adds the acquisition/materialization foundation. Original C/M replay, protected selection/index, reusable workflow, isolated signed origin, caller and consumer integration, protected adoption and the production security patch remain incomplete and gated. No active adoption or permission change is included. Core PR #786 remains draft and unmerged. Existing exception expiry and all later review, signing and CI requirements remain enforced. |
Why
The dev-first security patch in nold-ai/specfact-cli#786 cannot reach native Requirements tests because its subject graph has an expired dependency trust review. This prerequisite builds proof tools from a separate frozen official dependency closure while retaining the original native algorithms and all final subject audits.
What changed
Fixed context validation and acquisition keep candidate R, trusted core C, caller E, reusable policy P and the later protected index I distinct. Acquisition independently selects active I/P, authenticates exact Git blobs and canonical base64, verifies captured tool bytes, and rechecks main last. Caller E cannot select a policy revision, digest, URL, scope or credential.
The new materializer owns one private lifetime for the reviewed UV, Python and GitHub CLI assets, the hash-pinned 57-distribution tool environment, and exact C/M Git sources. It verifies archive and executable hashes, archive membership and links, API/Git identities, source bytes and modes, installer receipts and inventory. Subprocesses use owned output files, explicit environments, deadlines and child limits. Cleanup covers success and failure. Per-file limits and post-fetch aggregate checks are documented accurately; they are not an atomic total-disk quota.
Python is installed through pinned UV's explicit offline local-file mirror. Only that command permits manual installation; later UV commands prohibit interpreter downloads. The same frozen pip report records actual selected hashes for all final distributions. Git archives explicitly use tar.umask=0022 so strict source-mode validation accepts real Git output. No native engine, Git tree hasher, public fixture override or credential route was added.
Verification
9a872c1.46a4f141de4497c1f46a6114683308fd6f3b2525, all 102 focused materializer tests and all 1036 policy tests pass locally, with zero errors or skips. The existing hashed five-package test environment is unchanged.Remaining gates
Actual Ubuntu 24.04 x86_64 development run 40286288965, attempt 1, succeeded using this exact materializer at
46a4f141de4497c1f46a6114683308fd6f3b2525. Independent review verified the 57 installed tool distributions and their actual artifact hashes, all 4,555 command receipts, exact C/M source bytes, mandatory Linux cleanup controls, final deletion and transport restoration. Its single authenticated diagnostic artifact is12633554840. The four selection responses remain explicit development fixtures; this is no protected selection or native-origin authority. Conflicting runner image-revision labels are retained; no single reconciled image revision is claimed.Original C/M entrypoints, all fifteen intended assertion failures, and complete bounded process/named-path evidence still require the next independently reviewed replay unit. Unit-test, materialization and original-execution evidence remain separate.
The reusable workflow, isolated signed origin, caller permissions and producer/fresh/final consumers must be integrated and independently verified before a separate exact protected-deployment decision. Genuine fifteen-case native RED must precede the production security patch, followed by fresh final authority, final subject audits and all required CI.
The owner marked this PR ready for review on 26 October 2026 and authorized a normal merge into
nold-ai/.github:mainafter required checks, review findings and independent landing verification pass. The exact policy pin must remain in ancestry. Codex security review completed on the current46a4f141de4497c1f46a6114683308fd6f3b2525revision with no findings.This landing adds the reviewed acquisition/materialization foundation and frozen tool files. It does not activate a release index, add caller permissions, wire the reusable native workflow, establish authenticated native RED/origin, or complete the core security patch. Those follow-up gates above remain required before their corresponding adoption and core delivery. Core PR nold-ai/specfact-cli#786 stays draft until its complete patch and required verification are ready. No core merge, automatic merge, publication, local signing, additional protected activation/deployment, access expansion, check waiver or CVE exception renewal is authorized.