# Static Checks, Types and Builds — Safe Autonomous Code Fixing

Source: https://www.skillbyai.com/en/safe-autonomous-code-fixing/v-static

> Cheap checks catch whole classes of errors.

## Lint, type-check, build, scan

Beyond tests, run the project's **linters**, **type checkers** (mypy, TypeScript), **build**, and **security scanners** (dependency audit, static analysis) on every candidate. They catch unused variables, wrong types, unreachable code and risky patterns that tests may miss. Make the auto-fix pipeline use the same configuration as the project's CI so results agree, and never let a patch weaken lint or type settings to pass.

## A verification gate list

All must pass for a candidate to advance.

```text
gate                         example command                     pass condition
reproduction test            python -m pytest tests/test_x.py::t  passes (failed before)
full test suite              python -m pytest -q                  all pass
lint                         ruff check .                         no new findings
types                        mypy src                             no new errors
build                        python -m build / npm run build      succeeds
security                     pip-audit / npm audit, secret scan   no new high findings
scope                        diff stats + protected paths         within limits
```

## Compare against the base branch

Judge "no new findings" relative to the base branch so pre-existing warnings do not block every fix.

**Quiz:** What should happen if a patch disables a lint rule to pass?

- [ ] Accept it as a fix
- [x] Treat it as a protected change and reject it from automated fixing
- [ ] Prefer it over other candidates
- [ ] Ignore lint entirely

*Answer:* Treat it as a protected change and reject it from automated fixing. Do not weaken the checks.