Category

Review AI Agent Loops

Self-review and independent verification loops for safer AI-generated code changes. Every page is server-rendered, source-aware, and includes a copy-ready Markdown protocol.

All categories · 3 loops

All Review loops

Reviewmanual trigger Hardened

Independent Verifier Pass

Review an agent-produced change as a skeptical verifier before it is marked done.

CursorClaude CodeCodex

reviewverificationquality

Kickoff preview

Act as an independent verifier. Restate the requirements, inspect the diff, run relevant checks, and return a pass/fail/blocker verdict with evidence.
Goal: Prove the implementation satisfies the request without relying on the implementer’s claim.
Check command: git diff --stat && git diff --check
Exit condition: The verifier can map each requirement to code and verification evidence.
Max iterations: 3
Guardrails:
- Do not weaken, skip, delete, or rewrite the validation command to force success.
- Do not claim completion until the stated exit condition is actually satisfied.
- If blocked, report the blocker, evidence, and next safest action instead of gaming the metric.
- Do not rubber-stamp. Find at least one plausible failure mode and check it.

Reviewmanual trigger Hardened

PR Self Review

Have the agent review its own pull request for requirement coverage, risky changes, and missing tests.

CursorClaude CodeCodex

pull-requestreviewdiff

Kickoff preview

Review your own PR against the original task, changed files, tests, and risk categories. Fix must-fix issues and report evidence.
Goal: Identify problems in the agent’s own PR before human review.
Check command: git diff --stat main...HEAD && git diff --check main...HEAD
Exit condition: No must-fix review issues remain, or unresolved issues are clearly documented.
Max iterations: 3
Guardrails:
- Do not weaken, skip, delete, or rewrite the validation command to force success.
- Do not claim completion until the stated exit condition is actually satisfied.
- If blocked, report the blocker, evidence, and next safest action instead of gaming the metric.
- Do not hide unresolved issues; label them as risks or blockers.

Reviewmanual trigger Hardened

Regression Risk Audit

Map a change to likely regressions, then run or propose focused checks for those risks.

CursorClaude CodeCodex

riskauditcall-chain

Kickoff preview

Map the changed files to affected call chains, list likely regressions, verify the top risks, and report remaining uncertainty.
Goal: Identify likely regressions and verify the highest-risk paths.
Check command: git diff --name-only main...HEAD
Exit condition: Top regression risks are either tested, ruled out, or documented as unresolved.
Max iterations: 4
Guardrails:
- Do not weaken, skip, delete, or rewrite the validation command to force success.
- Do not claim completion until the stated exit condition is actually satisfied.
- If blocked, report the blocker, evidence, and next safest action instead of gaming the metric.
- Do not overstate confidence when a risk was not actually verified.

Why this category matters

Review loops are useful because AI agents need explicit feedback gates and exit criteria. A plain prompt may produce a first draft; a loop tells the agent how to verify progress, handle failures, and stop with evidence.