fix(review): strip leftover range interiors from posted suggestions

AGENTS.md

@@ -2,3 +2,9 @@
 
 <!-- CWL-ENTRY -->
 > **Agents: read the master context FIRST.** Before any work, read [`docs/CWL-MASTER-CONTEXT.md`](docs/CWL-MASTER-CONTEXT.md) (mission · naruon-as-platform + inter-component UML · cross-cutting disciplines · conventions · roadmap · current state), the live **GitHub Project #1** <https://github.com/orgs/ContextualWisdomLab/projects/1> (work/roadmap source of truth), the full spec **ContextualWisdomLab/naruon#974**, and operate the Project per [`docs/agent-github-project-protocol.md`](docs/agent-github-project-protocol.md). The repo/Project — not any private agent memory — is the source of truth.
+
+Materialize accepts only exact SHA-256 pins or a bounded relative `-r` include (no `.`/`..`); a lone `--require-hashes` directive is not trust evidence. `base_hash_locks` discovers hash-pinned `requirements/*.txt` via `_is_candidate_lock_path`. See [`docs/doctoring/review-inline-comment-422-fallback.md`](docs/doctoring/review-inline-comment-422-fallback.md) and [`docs/doctoring/trusted-uv-lock-materialization.md`](docs/doctoring/trusted-uv-lock-materialization.md).
+LEFT leftover suggestion fences are not applyable overview ranges.
+Leftover Manual-edit excerpts strip HTML comment delimiters and metacharacters.
+Failed-check diagnosis must pass `$control_json` to `build_inline_comment_failure_body`. Bare `422` substrings are not HTTP 422.
+

ARCHITECTURE.md

@@ -0,0 +1,48 @@
+# Architecture — ContextualWisdomLab/.github
+
+This repository is the organization-wide GitHub special repository: profile
+page, central PR governance / required workflows, and Cloudflare IaC. It is
+not naruon itself. naruon remains the email-workspace platform described in
+`docs/CWL-MASTER-CONTEXT.md`.
+
+```mermaid
+flowchart LR
+  subgraph Buyers["Sibling products"]
+    NAR[naruon]
+    ORCH[contextual-orchestrator]
+    BAND[bandscope]
+  end
+  subgraph Central["This repo"]
+    REV[OpenCode / Noema review]
+    STRIX[Required Strix]
+    SCH[Merge / autofix schedulers]
+  end
+  Buyers --> REV
+  Buyers --> STRIX
+  REV --> SCH
+  STRIX --> SCH
+```
+
+## Bounded requirement includes (2026-08-14)
+
+Coverage materialize treats a lone `--require-hashes` line as non-evidence.
+Only exact SHA-256 package pins or a bounded relative `-r`/`--requirement`
+include (`target == PurePosixPath.as_posix()`, no `.`/`..` parts, candidate
+lock path only) enter the trusted image. Dotted includes such as
+`./lock.txt` and `-r other-hashes.txt` stay outside the build context
+(CWE-22; CWE-1288).
+
+`base_hash_locks` discovers those candidates with `_is_candidate_lock_path`,
+so a hash-pinned `requirements/ci.txt` is materialized even though its
+file name is not `requirements*.txt`.
+
+## Failed-check review publication (2026-08-14)
+
+`run_failed_check_diagnosis` must pass the local `$control_json` into
+`build_inline_comment_failure_body`. The helper requires that third
+argument; a two-argument call under `set -u` aborts the publish step and
+discards a valid REQUEST_CHANGES diagnosis. Inline-comment 422 retry
+classifies HTTP 422 only from `HTTP 422` / `Unprocessable Entity` /
+classified GitHub HTTP 422 JSON phrases, not a bare `422` substring
+(CWE-1288). Decision record:
+`docs/doctoring/review-inline-comment-422-fallback.md`.

CHANGELOG.md

@@ -13,6 +13,43 @@ Semantic Versioning where the repository publishes a release.
 
 ### Fixed
 
+- `run_failed_check_diagnosis` now passes the local `$control_json` into `build_inline_comment_failure_body`, so a valid REQUEST_CHANGES failed-check diagnosis is not discarded by `set -u` when the helper requires the third argument.
+- `base_hash_locks` now discovers hash-pinned direct children such as `requirements/ci.txt` via `_is_candidate_lock_path`, not only `requirements*.txt` file names.
+- Inline-comment 422 retry now recognizes HTTP 422 only from `HTTP 422` / `Unprocessable Entity` / classified GitHub HTTP 422 JSON phrases. A bare `422` in a SHA, issue number, or run URL no longer starts the 20-comment retry loop (CWE-1288).
+- Materialized base Python locks only when every package line is an exact SHA-256 pin or a bounded relative `-r`/`--requirement` include. A lone `--require-hashes` directive, a dotted include such as `./lock.txt`, or `-r other-hashes.txt` no longer enters the trusted build context.
+- Posted review JSON now strips closed GitHub ` ```suggestion ` fences whose range overlaps leftover `path:start-end` coverage, so leftover `example.py:5-7` cannot still offer a one-click apply on interior `example.py:6` (CWE-1288).
+- Leftover Manual-edit excerpts now drop HTML comment delimiters and `<`, `>`, `&` before they enter the overview comment, so a leftover `-->` cannot close `<!-- opencode-review-overview -->` (CWE-116).
+- The leftover heading now prefixes the widest leftover `path:start-end` when a nested leftover range such as `example.py:5-6` sits inside leftover `example.py:5-7`, so authors see one leftover range then Manual-edit excerpts.
+- The leftover heading now prefixes one leftover `path:start-end` for leftover interiors of that range, so leftover `example.py:6` next to leftover `example.py:5-7` shows the range once then Manual-edit excerpts instead of two leftover reason bullets.
+- Leftover receipts now accept `path:start-end` as well as `path:line`, so a leftover cannot-provide range such as `example.py:5-7` drops overlapping applyable interiors (`example.py:6` and `example.py:5-7`) instead of being ignored.
+- The hunk-filter write path now omits applyable `path:start-end` rows from `--applyable-locations` when a leftover cannot-provide or LEFT line sits inside that range, so `applyable.txt` cannot list a one-click apply for the same span as leftover `example.py:6`.
+- The leftover overview CLI now omits applyable `path:start-end` rows that contain a leftover cannot-provide or LEFT line, so `--leftover-diff-locations` plus `--applyable-locations` show Manual-edit instead of a one-click apply for the same span.
+- Dropped applyable overview `path:start-end` rows when a leftover cannot-provide or LEFT line sits inside that range, so authors see the Manual-edit instead of a one-click apply for the same span.
+- Stopped listing LEFT-side leftover GitHub suggestion fences as applyable ranges, because GitHub cannot apply a suggestion on the deleted side.
+- Kept leftover Manual-edit excerpts on the overview CLI when `--leftover-diff-locations` sits inside a trusted `--deferred-locations` `path:start-end` even if that leftover line is not a trusted control finding.
+- Kept leftover Manual-edit excerpts that sit inside a trusted deferred multi-line `path:start-end` even when that exact leftover line is not itself a trusted finding, so authors still see the deferred range then the Manual-edit.
+- Grouped leftover Manual-edit excerpts that sit inside the same deferred multi-line `path:start-end` under one deferred range line, so authors do not see a repeated deferred prefix.
+- Omitted the duplicate leftover reason bullet when a leftover line sits inside a deferred multi-line `path:start-end`, so authors see the deferred range then the Manual-edit excerpt.
+- Omitted the duplicate leftover reason bullet when the leftover heading already prefixes a deferred range/origin for the same `path:line`, so authors see one deferred line then the Manual-edit excerpt.
+- Listed a deferred leftover ahead of its Manual-edit excerpt in the leftover heading when the same `path:line` is both deferred and leftover, so authors see the unposted fence first.
+- Kept both the leftover Manual-edit ` ```diff ` block and the deferred range/origin row when a cannot-provide or pure-deletion leftover sits past the 20-comment 422 retry cap, and still omitted those fences from applyable GitHub suggestions.
+- Recorded leftover OpenCode comments past the 20-comment 422 retry cap as deferred overview ranges with their LEFT origin, and stopped listing them under applyable GitHub suggestions because those comments are never posted.
+- Kept `start_line`/`start_side` on remapped leftover OpenCode suggestions when a batch 422 is retried one comment at a time, so a multi-line RIGHT range still posts as one GitHub suggestion.
+- Labeled remapped leftover OpenCode applyable ranges with the original LEFT `path:line` so the overview shows `path:right` came from LEFT `path:left`.
+- Anchored remapped leftover OpenCode LEFT comments to the first RIGHT line of the same `@@` hunk when the original line is gone, so multi-hunk files do not attach the suggestion to an earlier hunk.
+- Remapped leftover OpenCode LEFT suggested-diff comments onto a same-path current-head RIGHT hunk when one exists so those replacements become one-click GitHub suggestions instead of leftover manual-edit blocks. Pure deletions and cannot-provide fences stay leftover.
+- Persisted leftover OpenCode `cannot-provide` and `LEFT` suggested-diff replacement text as a distinct overview “Manual edit (not a GitHub suggestion):” ```diff block so authors can copy the change by hand without treating it as an applyable `path:line` / `path:start-end` GitHub suggestion.
+- Distinguished applyable OpenCode GitHub suggestion ranges from leftover ```diff fences (`cannot-provide` or `LEFT`) in the overview receipts so authors can see which hunks are one-click applies and which still need a manual edit.
+- Listed applyable OpenCode GitHub suggestion ranges (`path:line` or `path:start-end`) in the overview receipts so authors can see which surviving hunks shipped as one-click applies.
+- Set `start_line`/`line` on surviving multi-line OpenCode GitHub suggestions so a replacement that spans more than one current-head hunk line applies as one range.
+- Converted surviving OpenCode inline suggested diffs into GitHub `suggestion` blocks so authors can apply the replacement on the current-head hunk in one click.
+- Dropped OpenCode inline comments that sit outside every current-head changed hunk before the GitHub POST so those comments become overview receipts instead of a 422 that wipes the batch.
+- Capped one-at-a-time OpenCode inline retries at 20 comments and listed attached `path:line` beside refused receipts so the overview shows both outcomes, plus any locations left untried by the cap.
+- Kept each refused OpenCode inline comment's own GitHub 422 phrase next to its `path:line` so mixed retries do not collapse every failure into one shared error sentence.
+- After a mixed one-at-a-time inline retry, listed only the refused `path:line` rows in the overview receipts so attached hunks are not reported as failed.
+- After a batch GitHub 422, retried OpenCode inline comments one at a time so comments on surviving hunks still attach instead of dropping the entire review thread.
+- Stored each refused OpenCode inline comment as a durable overview receipt that pairs the trusted `path:line` with the GitHub 422 error phrase from `gh api` stderr or JSON `errors[].message`.
+- Named each trusted `path:line` in the OpenCode GitHub 422 inline-comment fallback so a refused attach still tells the author the exact current-head location instead of a generic “cited finding lines” sentence.
 - Bounded the Strix quality self-test's deterministic timeout fixtures to 3-second process and 5-second fake-sleep budgets so exact-head policy evidence completes inside the existing job limit without changing production Strix scanner timeouts, providers, credentials, or review semantics.
 - Allowed commas and ASCII parentheses in the bounded Strix changed-file path policy so legal tracked Packrat fixtures can receive exact-head security analysis, while rejecting raw `..` components before normalization and keeping controls, backslashes, whitespace ambiguity, and shell punctuation fail-closed.
 - Bound each review-agent invocation key to the wrapper's complete canonical payload, including the base branch and requesting actor; altered fields with a valid-format key now fail before durable-leader election or forwarding, and wrapper write permission is job-scoped.

CLAUDE.md

@@ -10,7 +10,11 @@ disciplines CP-1..CP-5/G6/SEAM, binding engineering conventions in §7, roadmap)
 [GitHub Project #1](https://github.com/orgs/ContextualWisdomLab/projects/1) (work/roadmap source of
 truth), and operate the Project per [`docs/agent-github-project-protocol.md`](docs/agent-github-project-protocol.md).
 The repo/Project — not private agent memory — is the source of truth. This file complements those
-documents; it does not replace them.
+documents; it does not replace them. Materialize accepts only exact SHA-256
+pins or a bounded relative `-r` include; `base_hash_locks` discovers
+hash-pinned `requirements/*.txt` via `_is_candidate_lock_path`. Bare `422`
+substrings are not HTTP 422. `run_failed_check_diagnosis` must pass
+`$control_json` to `build_inline_comment_failure_body`.
 
 ## What this repository is