fix: drop references to the removed docs/topics tree

[kyle-sexton]

No related issue: main has been red since c11c7ec (#650); this restores it.

Summary

#650 deleted docs/topics/ while #656 added two AGENTS.md links into it, and two tests still read files there. lychee-offline and selector-contract fail on main and on every open PR.

Fix

• AGENTS.md: the two Code Review Rules lines that linked into docs/topics/ now state their rule inline.
• .github/scripts/ci-fanout-consolidation.test.cjs: drop the ADR read and the test that asserted the ADR's text.
• .github/scripts/claude-review-plugin-path.test.cjs: drop the test that asserted the V2 architecture doc's text.

Verification

• node --test .github/scripts/*.test.cjs: 192 pass, 0 fail.
• git grep docs/topics finds only the standards-managed .github/actionlint.yaml comment that chore: remove docs/topics #650 already noted.

Related

• Unblocks feat(claude-lanes): add the intake-triage reusable lane #659.
• Cause: chore: remove docs/topics #650 and docs: add Code Review Rules section for Codex review #656 merged in parallel.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-03T19:06:42.022198Z	15557a0	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[claude]

Claude finished @kyle-sexton's task in 2m 10s —— View job

---

Code review complete

• Gather context (diff, changed files, PR body)
• Review AGENTS.md changes
• Review test file changes (ci-fanout-consolidation.test.cjs, claude-review-plugin-path.test.cjs)
• Verify claims (grep for docs/topics)
• Post findings

Summary

The fix itself is correct and minimal: it removes the two now-dangling docs/topics/... links from AGENTS.md and the two test assertions that read files under the deleted docs/topics/ tree, matching exactly what #650 broke. I independently confirmed via git grep/Grep that the only remaining docs/topics reference repo-wide is the already-noted comment in .github/actionlint.yaml, and that neither removed doc path (ci-fanout-consolidation/ADR.md, claude-review-lanes/V2-PLUGIN-ARCHITECTURE.md) nor local-lane-guards.md is referenced anywhere else. Both edited test files end cleanly with no leftover unused imports or dangling syntax.

I could not run node --test .github/scripts/*.test.cjs myself in this sandbox (command execution was blocked), so the PR's "192 pass, 0 fail" claim is author-reported and unverified by me — the static review above is what I was able to confirm independently.

One non-blocking documentation-consistency finding posted inline on AGENTS.md (lines 13-14): the file's own preamble ("the linked file states it in full") no longer holds for the two rewritten bullets, which now state their rule inline with no link. A home for at least one of them already exists in README.md.

[claude]

Claude finished @kyle-sexton's task in 50s —— View job

---

Security review: no issues found

• Gather context (PR diff, changed files)
• Review changed files for security-relevant issues
• Check AGENTS.md instruction-surface deletions against protected classes
• Post findings

Scope: AGENTS.md, .github/scripts/ci-fanout-consolidation.test.cjs, .github/scripts/claude-review-plugin-path.test.cjs — all consistent with the already-merged docs/topics removal in #650.

Instruction-surface check (AGENTS.md): the two edited lines each drop a markdown link into the now-deleted docs/topics/ tree, but the rule statement itself is unchanged word-for-word in both cases — "Local-lane guard wrappers keep parity with the standards component" and "Lanes consolidate as composite actions; ci-status stays the single required check." No enforcement content, scope, or exception was narrowed; only a dead pointer to a file #650 already deleted was removed. This does not match a protected-class removal — nothing a hook, allow/deny rule, or validator relied on changed. Note: docs/conventions/instruction-exception-register/README.md is not present in this checkout, so the protected-class list itself was not consulted directly; the consequence-based check above stands in for it.

Test deletions: both removed test(...) blocks only asserted string content of docs files that no longer exist (the ADR and the V2 architecture doc) — no behavioral or security assertions were removed, and no fs.readFileSync calls were left dangling on paths outside the repo tree.

No CRITICAL, IMPORTANT, or SUGGESTION findings. This is a straightforward doc/test cleanup PR with no security-relevant surface.

[github-actions]

claude-security-review has reviewed this pull request through 1b6e287; a later push is reviewed from there.

[github-actions]

claude-review has reviewed this pull request through 1b6e287; a later push is reviewed from there.

[claude]

Claude finished @kyle-sexton's task in 1m 14s —— View job

---

Incremental code review (AGENTS.md only)

• Read .claude-lane/incremental.diff (the only change since 15557a0)
• Read current AGENTS.md
• Check whether the change resolves the inconsistency flagged in the prior review
• Post findings

Summary

Scope for this pass: only the line changed since 15557a0 — AGENTS.md:3, where 8cdfe5c reworded the Code Review Rules preamble from "the linked file states it in full" to "a linked file states it in full."

One finding posted inline on AGENTS.md:3: the article swap doesn't actually close the gap the prior review raised. The preamble still implies every line has a linked file that states its rule in full, but the two bullets on lines 13-14 carry no link at all, so neither "the" nor "a" linked file exists for them. A real fix is still either linking line 13 to README.md's existing local-lane-guards paragraph, or explicitly carving out inline-only bullets in the preamble's wording.

Not blocking — the PR's actual goal (unbreaking lychee-offline/selector-contract) is unaffected by this wording.

[claude]

Claude finished @kyle-sexton's task in 48s —— View job

---

Security review: no issues found

• Gather context (incremental diff since 15557a0, scoped to AGENTS.md)
• Run /review:security-review per the lane's incremental scope
• Post findings

Scope: this lane already reviewed the PR through 15557a0. The only change since then is one line in AGENTS.md:

-Each line names a rule CI does not enforce; the linked file states it in full.
+Each line names a rule CI does not enforce; a linked file states it in full.

Instruction-surface check: this sentence is the file's own meta-description of its bullet-list convention, not a standing instruction that itself gates an action elsewhere in the tree (no hook, allow/deny rule, or validator keys off its wording). Changing "the" to "a" only acknowledges what the prior commit (15557a0, already reviewed) had already done — stating two rules inline instead of via a link — and was the exact fix the earlier review comment on this PR suggested ("either the preamble needs to be loosened to admit inline-only rules, or that statement needs a new home"). No rule's enforcement, scope, or exception changed; no rule text was removed or softened. This does not match a protected-class removal under the instruction-exception-register lens. Note: docs/conventions/instruction-exception-register/README.md is not present in this checkout, so the protected-class list itself was not consulted directly; the consequence-based check above stands in for it.

No CRITICAL, IMPORTANT, or SUGGESTION findings.

[claude]

Claude finished @kyle-sexton's task in 56s —— View job

---

Security review: no issues found

• Read .claude-lane/incremental.diff (the only change since 8cdfe5c)
• Run /review:security-review per the lane's incremental scope
• Post findings

Scope: this lane already reviewed the PR through 8cdfe5c. The only change since then is the Code Review Rules preamble in AGENTS.md:

-Each line names a rule CI does not enforce; a linked file states it in full.
+Each line names a rule CI does not enforce. A line with a link points at the file that states the
+rule in full; a line without one states the whole rule itself.

Instruction-surface check: this sentence is the file's own meta-description of its bullet-list convention — not itself a standing instruction that gates an action elsewhere in the tree (no hook, allow/deny rule, or validator keys off its wording, consistent with what the two prior reviews on this PR already found for the earlier "the"→"a" wording tweak). This revision doesn't remove, narrow, or soften any rule's enforcement, scope, or exception: it only reconciles the preamble with what 15557a0 had already done two commits earlier (stating the two guard/consolidation rules inline instead of via a now-deleted docs/topics/ link). That inline content is unchanged word-for-word across this whole PR. This is in fact the fix both prior review comments on this PR requested — explicitly admitting inline-only bullets rather than relying on an article swap to imply it — so it closes the gap rather than opening one. No protected-class removal under the instruction-exception-register lens. Note: docs/conventions/instruction-exception-register/README.md is not present in this checkout, so the protected-class list itself was not consulted directly; the consequence-based check above stands in for it.

No CRITICAL, IMPORTANT, or SUGGESTION findings.

[claude]

Claude finished @kyle-sexton's task in 37s —— View job

---

Incremental code review (AGENTS.md only)

• Read .claude-lane/incremental.diff (the only change since 8cdfe5c)
• Read current AGENTS.md
• Check whether the change resolves the inconsistency flagged in the two prior reviews
• Post findings

Summary

Scope for this pass: the one change since 8cdfe5c — AGENTS.md:3-4, where 1b6e287 rewrote the preamble from:

Each line names a rule CI does not enforce; a linked file states it in full.

to:

Each line names a rule CI does not enforce. A line with a link points at the file that states the
rule in full; a line without one states the whole rule itself.

This closes the gap both prior reviews on this PR raised: it explicitly carves out inline-only bullets rather than implying every line has a backing linked file. The two unlinked bullets this preamble now has to cover — AGENTS.md:14 ("Local-lane guard wrappers keep parity with the standards component.") and AGENTS.md:15 ("Lanes consolidate as composite actions; ci-status stays the single required check.") — each state their rule wholly inline with no link, matching the new wording exactly.

No findings on this increment. The earlier suggestion to instead link line 14 to README's local-lane-guards paragraph was one of two valid options the prior review offered; the author took the other (loosen the preamble), which is internally consistent and resolves the issue.