Code Change Verification
Purpose
Do not mark an execution task as complete until the repo’s verification stack has passed.
This skill is intended to reduce regressions by making “run the checks” the default behavior for agents when they change runtime code, tests, build config, or deployment scripts.
When to run
Run the full stack when you touch:
- Runtime code
- Auth/billing flows
- Build/test tooling
- CI scripts, lint config, or typecheck config
You may skip the full stack for:
- Docs-only changes
- Comments-only changes
- Formatting-only changes that do not affect runtime behavior
If unsure, run it.
What to run (policy)
- First, follow any repo-local requirements (e.g.
AGENTS.md, runbooks). - Then run the repo’s standard verification commands (format → lint → typecheck → tests/build).
- If anything fails: fix the issue and re-run the full stack.
Behavioral heuristic changes (required):
- Follow the canonical review-comment doctrine in
AGENTS.md. - If you change parser, selector, routing, or guard logic, verification must include:
- the exact reported failure,
- at least one adjacent valid input,
- at least one adjacent false-positive input.
- Fail verification if the patch only appeases reviewer wording or narrows fixtures/contracts without an intentional documented contract change.
This repository (agentic-cockpit)
Run:
npm test
Optional (if configured in the future):
npm run lint
Helper scripts (recommended)
- macOS/Linux:
bash .codex/skills/code-change-verification/scripts/run.sh - Windows:
powershell -ExecutionPolicy Bypass -File .codex/skills/code-change-verification/scripts/run.ps1
If the script cannot detect a supported toolchain, fall back to the repo’s documented commands and
report what you ran in testsToRun.
Learned heuristics (SkillOps)
Converted and distributed by TomeVault — claim your Tome and manage your conversions.