paulund/quality-gate

Quality Gate

Run the project's static-checks + build sequence. Stop on the first failure, diagnose the root cause, and fix it before re-running.

Workflow

Step 1 — Detect the toolchain

Read package.json scripts (or equivalent for the project: composer.json, Makefile, Cargo.toml). Pick the first command from each row that exists:

| Phase | Preferred | Fallback | |---|---|---| | Lint | pnpm lint / npm run lint | npx next lint, composer run lint | | Types | pnpm types / npm run types | npx tsc --noEmit | | Tests | pnpm test / npm test | composer run test, framework default | | Build | pnpm build / npm run build | npx next build |

If a phase has no command for the project (e.g. a JS-only repo with no types script), skip that phase and continue.

Step 2 — Run in order

Run the four phases sequentially. Each must pass before the next runs.

<lint command>
<types command>
<tests command>
<build command>

Step 3 — On failure, Stop-the-Line

1. STOP running later phases.
2. PRESERVE the full error output — don't truncate it from your context.
3. DIAGNOSE — reproduce the failure if needed, trace the root cause, not the symptom. Use git log / git blame / existing tests for context.
4. FIX the root cause.
5. GUARD — if the failure was a regression, add a test that would have caught it.
6. Re-run the full gate from the top.

Max 2 diagnose-fix cycles. If still failing, stop and surface the failure to the caller — don't hide it.

Step 4 — Report

On success: one-line confirmation listing the phases that ran (lint, types, tests, build). On failure: the failing phase, the command that ran, and the first ~20 lines of error output.

Constraints

MUST DO

• Detect toolchain commands from project files; do not assume pnpm everywhere.
• Run phases in the documented order — fast feedback first.
• Surface failure output verbatim to the caller; never silence it.
• Add a regression test on Step 3 step 5 when the failure is a regression.

MUST NOT DO

• Skip a phase to "save time" — the gate is the contract.
• Run later phases after an earlier one failed.
• Loop more than 2 diagnose-fix cycles before stopping.
• Mask failures by editing the test or config to make them pass without fixing the underlying issue.