impact-review
Skill16 starsChanged 39 days agoMITView on GitHub
Run the review→fix→test loop on a change, scoped to its blast radius: map impact, review through 3 lenses in parallel, adversarially verify each finding, fix only confirmed ones, test the affected area, then the QA gate. Bounded retries. (H3/H7).
In this repository16 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- .github/copilot-instructions.mdCopilot instructions
- analyze-requirementsSkill
- be-fe-security-auditSkill
- commit-deploy-logSkill
- deep-researchSkill
- deploy-to-testSkill
- evidence-bundleSkill
- fanSkill
- fix-and-verifySkill
- implement-changeSkill
- qa-gateSkill
- run-affected-testsSkill
- verify-implementationSkill
- verify-requestSkill
The file
Files
impact-review
The Agent Pack's orchestration skill — the review→fix→test loop that all nine
2026-08-03 consuming-project proposals independently designed. It ties together
the four new Agent Pack agents plus the existing qa-gate, and its whole design
principle is narrow but safe: work only on the change's blast radius (cheap,
fast) without ever under-scoping in a way that ships a bug.
Three safety chokepoints, in order, because each catches what the previous cannot:
- Adversarial verify before any fix — a plausible-but-false finding never reaches the fixer, so the loop cannot introduce the bug it was meant to catch.
- Bounded retries —
max_fix_retriesfromcasan-policies.yamlcaps the fix↔test loop; hitting the cap escalates to a human rather than looping. - qa-gate re-runs the suites — the final gate does not trust the tester's report; it re-executes, and an unrun suite yields CHANGES_REQUIRED, never APPROVED.
Impact selection and the test runner come from project-owned config
(contracts/impact-map.yaml, .harness/control/agent-config.yaml); the skill and
agents ship in the bundle. That two-layer split (C2) is what lets one common
pipeline serve a Vitest repo, a pytest repo and a PHPUnit repo without hardcoding
any of them.
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.

