Cold Pr Review Until Clean

Cold PR review until clean

Run independent cold-review subagents in a loop. Every time a cold reviewer surfaces actionable findings, fix only those findings and run another fresh cold-review subagent.

Stop after one fresh cold review invocation completes with zero actionable findings.

This skill is the normal-PR final independent review loop. In code-review, run it after the Codex review phase unless Codex review is unavailable or explicitly skipped. The source of truth here is cold-pr-review, not an OpenClaw-specific review workflow or Clawsweeper. Fixes are handled directly by the implementing agent unless a repo-specific fix workflow applies.

Non-negotiables

review_tool: must invoke cold-pr-review through an independent subagent whenever the harness supports subagents
review_context: subagent gets only the target and neutral review checklist; no prior rationale or findings
fix_tool: apply targeted fixes directly, or use the repo-specific fix workflow when one exists
state_store: keep findings, commands, open queue, and stop reason in the findings CLI
scope_gate: inherit the persisted scope baseline; run scope-check after every accepted fix and stop immediately on non-zero
stop_condition: one cold review run with zero actionable findings
counter_reset: any actionable finding resets consecutive_clean to 0
no_early_exit: do not stop before a fresh cold review returns clean
no_self_review: do not substitute the implementer's judgement for a cold review
fresh_reviewer: use a new isolated subagent for each review pass whenever the harness supports it
consult_findings: consult-worthy findings go to the consult queue; keep fixing other findings instead of waiting
queue_matched_passes: a pass whose only findings match the open consult queue counts toward the clean target but can never produce a final clean verdict
fixed_point: when the clean target is met and the consult queue is non-empty, suspend as blocked-on-consult; never keep re-running reviews on an unchanged tree

Workflow

1. Pre-flight the target.

   Confirm the PR number, URL, branch, or git range and the exact committed HEAD. Refuse staged, unstaged, or untracked changes. Load review-guardrails and identify required verification commands. If running inside code-review, inherit the orchestrator's persisted scope budget, consult queue, and queue-matching rules. Confirm it with scope-status before fixing anything.

2. Build neutral reviewer context.

   If code-review already ran review-flow-map, pr-rubbish-audit, typescript-discipline, improve-codebase-architecture, reducing-cognitive-load, frontend-ui-validation, or finding-discipline, do not pass those results to the reviewer. Convert them only into neutral checklist topics.

3. Dispatch a fresh independent reviewer.

   Load wait-efficiently, then read references/subagent-dispatch.md. Done when a fresh isolated reviewer receives only the target and neutral review checklist, and the coordinator uses the native event-driven wait for its result. Match candidates against the findings registry only after the pass.

4. Run the until-clean loop.

   Read references/loop.md. Maintain consecutive_clean and iterations. Triage findings, fix actionable findings, record state in the findings CLI, and rerun with fresh reviewers until the clean target or an honest stop condition is reached.

5. Classify clean and report.

   Read references/clean-criteria.md before counting a pass clean. Read references/fixing-and-reporting.md before editing or reporting.

Done means

• Every review pass used a fresh isolated reviewer whenever the harness supported one.
• The reviewer did not receive prior findings, fixes already attempted, design rationale, CI confidence signals, desired verdicts, or earlier code-review results.
• One fresh cold review run completed with zero actionable findings on the reviewed tree, or the loop stopped honestly with blocked-on-consult or budget-expired.
• Findings, fixes, validation commands, consult-queue changes, and stop conditions are recorded in the findings CLI.
• Every accepted fix is followed by a passing scope-check, and the final scope status is not missing or blocked.
• Accepted fixes from one pass are committed together before the next reviewer.
• No code was edited between clean passes.
• No final clean verdict is reported while the consult queue has open entries.

Avoid

• substituting the implementer's judgment for a cold review;
• leaking prior findings or implementation rationale into reviewer prompts;
• reviewing staged, unstaged, or untracked changes;
• stopping without a fresh clean cold-review pass;
• silently fixing or rejecting consult-worthy findings;
• running more reviews on an unchanged tree beyond the clean target.