freeze-review-loop
Skill1 starsChanged yesterdayMITView on GitHub
Use when a frozen spec (frozen: true) must reach pass before an Auto build phase may start. Invoke explicitly: /freeze-review-loop or paste the prompt below. Not automatic by default — the operator or handover must request it. Complements cursor/skills/freeze-review/SKILL.md (single-pass review). This skill runs bounded rounds until pass, escalation, or max rounds — without re-pasting the prompt each round. Prevent dishonest or shallow freeze approvals. A thinking-model reviewer must cite file+line findings; a fixer applies only those citations; the reviewer…
In this repository28 other files it gives its agents.
- AGENTS.mdAGENTS.md
- .cursor/rules/build-verification-required.mdcCursor rule
- .cursor/rules/check-ok-thinking.mdcCursor rule
- .cursor/rules/governance-sync.mdcCursor rule
- .cursor/rules/l1-checkpoints-consumer-owned.mdcCursor rule
- .cursor/rules/no-docs-only-pr-to-main.mdcCursor rule
- .cursor/rules/orchestrator.mdcCursor rule
- .cursor/rules/print-next-closeout.mdcCursor rule
- .cursor/rules/tier-authority.mdcCursor rule
- .cursor/rules/workspace-authority.mdcCursor rule
- build-verification-reviewSkill
- check-okSkill
- deploy-verification-reviewSkill
- freeze-reviewSkill
- governance-syncSkill
- independent-second-reviewerSkill
- print-nextSkill
- workspace-statusSkill
- build-verification-reviewSkill
- check-okSkill
- deploy-verification-reviewSkill
- freeze-review-loopSkill
- freeze-reviewSkill
- governance-syncSkill
- independent-second-reviewerSkill
- knowtationSkill
- print-nextSkill
- workspace-statusSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 1 is distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 3
Use when a frozen spec (frozen: true) must reach pass before an Auto build phase may start.
- Very specific: line 19
Before clearing {step}b (Auto) in the handover
- Very specific: line 26
Read policy/model-labels.yaml and .overseer/config.yaml → freeze_contract.reviewer.
- Distinctive: line 80
After fixer commits, before next review round:
- Very specific: line 108
If the skill cannot run (no thinking model, context limit): run single-pass freeze-review skill + manual rounds in fresh chats.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Freeze-review loop skill
Use when a frozen spec (frozen: true) must reach pass before an Auto build phase may start.
Invoke explicitly: /freeze-review-loop or paste the prompt below. Not automatic by default
— the operator or handover must request it.
Complements cursor/skills/freeze-review/SKILL.md (single-pass review). This skill runs bounded
rounds until pass, escalation, or max rounds — without re-pasting the prompt each round.
Purpose
Prevent dishonest or shallow freeze approvals. A thinking-model reviewer must cite file+line findings; a fixer applies only those citations; the reviewer re-checks until clean or stopped.
When to use
- Roadmap step
{step}a(Thinking) just committed a freeze artifact ok review --freeze <path>returnedfindingsor you need deep review beyond the checklist- Before clearing
{step}b(Auto) in the handover
Model
- Reviewer rounds:
thinking-highlabel — extended thinking / Opus-class in Cursor - Fix rounds: may use Auto for mechanical citation fixes only; reviewer always re-runs thinking
Read policy/model-labels.yaml and .overseer/config.yaml → freeze_contract.reviewer.
Hard stops (stop loop immediately)
Escalate to human — do not auto-fix — when any finding is:
security— auth, scope, secrets, injectionirreversible— data deletion, non-revertible migrationsreal_money— billing, live spendgates_tier3— artifact gates merge, staging push, live flip
Also stop when:
- Verdict is
blocked - Round count reaches max rounds (default 7)
- Fixer would need to change scope beyond cited lines (Tier 2 — ask once)
Loop (execute in one agent session — do not stop after one round)
ROUND = 1
MAX_ROUNDS = 7 # operator may lower to 3 for small specs
while ROUND <= MAX_ROUNDS:
1. REVIEW (thinking model, independent posture)
- Read frozen artifact + frozen_inputs it depends on
- Read docs/OVERSEER-KIT-SPEC.md §6
- Emit verdict: pass | findings | blocked
- Every finding: path:line, severity, category, concrete message
- Run checklist mentally: ground-truth edge, completeness, consistency,
security, irreversibility, real money, Tier-3 linkage, citation discipline
- Do NOT trust prior round summaries — re-read files
2. If pass:
- Run: ok review --freeze <artifact-path>
- If CLI pass: stamp ok; update artifact Review-record table; EXIT success
3. If blocked OR escalating finding:
- Record in Review-record table; STOP for human
4. If findings (non-escalating):
- FIXER: apply minimal edits addressing ONLY cited path:line items
- No scope expansion; no deleting requirements to greenwash
- Commit on feature branch (Tier 1)
- ROUND += 1; continue loop
If ROUND > MAX_ROUNDS:
- Verdict: blocked (max rounds)
- STOP for humanCLI between rounds (recommended)
After fixer commits, before next review round:
./cli/ok review --freeze <artifact-path> --dry-runChecklist gate catches structural gaps; thinking review catches semantic dishonesty.
Review-record table (required)
Append each round to the artifact's freeze Review record section:
| Round | Reviewer | Verdict | Resolution |
| N | Freeze-review loop (thinking) | findings | F1 fixed in … |
Must cite
Uncited findings are invalid — discard and re-review. Operator verifies citations, not trust.
Tier authority
- Feature-branch fixes during loop: Tier 1
- Merge to
{{vcs.git.main_branch}}: Tier 3 — never part of this loop - Enabling Cursor Automation for this: Tier 2 confirm once
Degrade
If the skill cannot run (no thinking model, context limit): run single-pass freeze-review skill +
manual rounds in fresh chats.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

