Plan Review Approval Shortlist Audit

Plan-Review Approval Shortlist Audit

Use when the user asks which status:plan-review issues are actually ready, closest to ready, or should be surfaced for approval next.

Why this exists

A naive queue sweep will misclassify issues if it relies only on:

• the presence of a status:plan-review label
• the newest review artifact filename
• a single overnight MINOR or approval-ready (conditional) summary

In practice, the real blocker state often lives across several surfaces at once.

Required audit surfaces

For each candidate issue, check all of these:

1. Live GitHub issue state

• current status:* labels
• issue still open
• recent comments for rollback notes, blocker updates, or recommendation caveats

2. Canonical local plan state

• docs/plans/YYYY-MM-DD-issue-NNN-*.md
• docs/plans/README.md row status

3. Review evidence

• latest canonical artifact under scripts/review/results/*-plan-NNN-*.md
• whether the artifact is real provider review vs placeholder/fallback
• whether a newer GitHub comment sharpens or contradicts the artifact conclusion

4. Approval evidence

• .planning/plan-approved/NNN.md
• absence of this marker means the item is still only a recommendation candidate, not approved
• First verify that the checkout used for marker inspection is not stale or behind the target branch. If the active worktree is behind origin/main, either fast-forward/use a fresh worktree or inspect the committed remote tree directly (for example git ls-tree -r --name-only origin/main -- .planning/plan-approved/NNN.md) before reporting the marker as missing. Stale governance worktrees can otherwise misclassify truly committed approval markers.

5. Explicit prerequisites

• machine constraints (machine:* labels)
• blocked-on-other-issue or blocked-on-artifact conditions
• prerequisites documented in comments or review artifacts (e.g. unreadable source PDFs, missing summaries, provider rerun still required)

Classification buckets

A. Clean approval candidate

Use only when:

• canonical plan exists
• review coverage is present and recent
• no fresher comment says needs-revision / not ready for user approval
• no unresolved prerequisite blocker is still active
• issue is still in status:plan-review
• no local approval marker yet

B. Best available approval candidate with caveat

Use when:

• item is stronger than the rest of the queue
• latest review is MINOR or approval-ready (conditional)
• but a residual condition still exists (for example specific machine requirement)

State the caveat explicitly. Do not silently upgrade to “cleanly approval-ready.”

C. Needs revision

Use when:

• latest or recent review/comments still record MAJOR, needs-revision, or unresolved review findings
• approval-readiness depends on provider reruns that have not happened yet
• plan quality may be high, but the review/governance standard has not yet been met

D. Governance drift / not actually in plan review

Use when:

• live status:plan-review exists but there is no canonical plan file
• or there are no review artifacts yet
• or the queue state implies a maturity level the artifacts do not support

Decision rules

• MINOR or approval-ready (conditional) is not enough by itself.
• Recent issue comments can overrule simplistic artifact parsing if they document fresher blocker state.
• A best-in-queue recommendation is not the same as certifying all blockers are gone.
• Never describe an item as already approved unless both live state and local approval evidence support that claim.

Output format

For a shortlist, return:

1. Approve now

• issue number + title
• why it is the strongest candidate
• residual caveat, if any

2. Keep in revision lane

• issue number + title
• 3-6 bullet punch-list of remaining blockers

3. Optional queue note

• whether the queue still contains governance drift items that should be cleaned before more approval triage

Practical pattern

If you need a fast but trustworthy pass:

1. enumerate live status:plan-review issues
2. separate missing-plan / missing-review drift first
3. inspect latest review artifacts for the remaining set
4. inspect recent GitHub comments for contradiction or conditionality
5. verify no local approval marker exists
6. produce a small shortlist with caveats instead of overclaiming certainty