land: accept a bot clean-review comment (naming the current head SHA) as the silence signal — fn-65

[gmickel]

Spec: fn-65-land-accept-a-bot-reviewed-clean · Plan-review: SHIP (rp, 3 rounds) · Completion-review: SHIP (rp) · Found dogfooding the fn-64 land.

What & why

/flow-next:land's silence merge signal needs "an automated review of the current head." But land detected reviews only via the formal reviews API — and Codex (the bot silence was built for) files a formal review only when it has findings. On a clean pass it posts an issue comment ("Didn't find any major issues. Reviewed commit <sha>") that never hits the reviews API. So a converged-clean PR could sit at NEEDS_HUMAN forever — which is exactly what happened landing fn-64 (PR #176); a human had to merge on judgment.

This teaches silence to also accept a bot's clean-review comment that names the current head SHA as head-current review evidence — closing the blind spot for the default bot-review flow.

R-ID coverage

R-ID	What	Where
R1	Head-current automated-reviewer clean comment satisfies silence	workflow.md §2.6 comment scan
R2	Stale-SHA or no-SHA comment never satisfies — explicit token extraction ([0-9a-fA-F]{7,40}, ≥7, non-empty, prefix-match), no ==$var*-on-empty footgun	workflow.md §2.6
R3	Non-automated login ignored (reuses the [bot]/automatedReviewers allowlist)	workflow.md §2.6
R4	Never bypasses an unresolved thread or red CI; scan is a read-only GET (dry-run-safe)	workflow.md §2.6
R5	Config contract: seeded structured built-in ERE; null→default, explicit ""→disabled; silence-gated; only ever sets AUTO_REVIEW_CURRENT	flowctl.py get_default_config() (+ dogfood copy)
R6	Docs: flowctl.md row + flow-next.dev land.mdx (3 spots) + changelog	docs/flowctl.md, flow-next.dev
R7	Observability: AUTO_REVIEW_SOURCE=comment + AUTO_REVIEW_EVIDENCE in dry-run/report	workflow.md §2.6

Critical changes — where to look

• plugins/flow-next/skills/flow-next-land/workflow.md §2.6 — the comment scan: paginated issues/<n>/comments, gated on REVIEW_SIGNAL == silence, runs after the reviews loop and before the draft-trigger check; allowlist + pattern + head-current-SHA-token match; sets AUTO_REVIEW_CURRENT=1 (never resets). SHA preferentially read from the Reviewed commit marker line.
• plugins/flow-next/scripts/flowctl.py (+ dogfood .flow/bin/flowctl.py, lockstep) — land.cleanReviewCommentPattern seeded as a structured ERE; the cfg read distinguishes null (→ built-in default) from explicit "" (→ disabled) — the real off-switch.
• plugins/flow-next/tests/test_land_config.py — +16 tests (default present, round-trip, explicit-empty-disables distinct from default, no-clobber both directions, docstring) plus a static §2.6 invariant guard (--paginate, silence-gate, before-draft-trigger ordering, SHA non-empty/min-length, set-only-never-reset, null≠empty).
• Codex mirror regenerated; flowctl.md + flow-next.dev land.mdx + changelog; plugin 2.1.0 → 2.1.1 (additive/opt-in patch). flow-next.dev pushed separately (gmickel/flow-next.dev@57f61da).

Decision context

silence was built for "bots that comment but never APPROVE," yet its detection only read formal reviews — so the feature and its implementation disagreed on exactly the bot it targets. The fix stays conservative: allowlist-gated, head-SHA-anchored (a fix push still forces a fresh clean comment), additive-only (never overrides a formal review / open thread / red CI), and disable-able via empty pattern.

Verification

• Full suite green (python3 -m unittest discover -s plugins/flow-next/tests); ci_test.sh 67/67; standalone behavioral bash matrix (head-current clean ✓, stale-SHA ✗, no-SHA ✗, non-bot ✗, empty-pattern ✗).
• Plan-review (RepoPrompt): SHIP after 3 rounds — caught the ==$var*-on-empty footgun and the un-disableable-opt-out config bug.
• Completion-review (RepoPrompt): SHIP — all R1–R7 Met.

🤖 Generated with Claude Code via /flow-next:make-pr

[gmickel]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Already looking forward to the next diff.

Reviewed commit: d56d5f80c0

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".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".