smoke-test-and-repair

Reproduce a CI failure locally by running the same gates in order: reproducible install (npm ci), typecheck, lint, build, then tests. Match the CI environment for Node version and filesystem case sensitivity, since case-sensitive imports often pass on macOS but fail on Linux CI.

Run gates in dependency order: install, typecheck (tsc --noEmit), lint, build, unit tests, then startup and route health checks. Stop at the first failing gate, fix the smallest root cause, re-run only that gate, then continue down the chain.

This usually traces to environment differences: filesystem case sensitivity (macOS vs Linux), stale local node_modules versus a clean npm ci in CI, path alias mismatches, or Node version drift. Reproduce with a clean install and matching Node version rather than adding suppressions.

Yes. It detects pnpm-workspace.yaml, workspaces fields, turbo.json, and nx.json, then uses the right aggregate commands like pnpm -r build or turbo run build. Failing packages are isolated with filters such as pnpm --filter or turbo run test --filter before fixing.

Skip a gate only when the script does not exist, a required secret like DATABASE_URL is unavailable, or running it would need a deploy or destructive action. Every skip must carry a reason and never counts toward an all-green verdict.

Adding @ts-ignore or eslint-disable masks a real defect instead of fixing it, hiding bugs that resurface at runtime. The correct approach is fixing the root cause at the lowest wrong layer, such as narrowing a type at its source rather than casting with as any.