Quality Gates Runner

Quality Gates Runner

Run quality checks via subagents to keep main context clean.

When to Use

• After implementation is complete
• Before creating a PR
• In target repository directory

Prerequisites

1. In target repository directory (the repo being worked on)
2. Changes exist (unstaged or staged)

Workflow

1. Verify Directory

Confirm in target repository and check for changes:

pwd  # Should be in target repository
git status --short  # Should show modified files

2. Dispatch Subagents in Parallel

Create tasks for each quality gate using the Task tool. Dispatch all in parallel:

Task 1: Lint

Run the lint command (e.g., `pnpm lint` or per AGENTS.md) in the target repository.
Report: PASS or FAIL with specific errors and file:line locations.
If FAIL, suggest specific fixes for each error.

Task 2: Format

Run the format check command (e.g., `pnpm format:check`) in the target repository.
Report: PASS or FAIL.
If FAIL, run the format fix command and report files changed.

Task 3: Typecheck

Run the typecheck command (e.g., `pnpm typecheck`) in the target repository.
Report: PASS or FAIL with specific type errors and locations.
For each error, suggest the fix.

Task 4: Unused Code Check (if available)

Run unused code detection (e.g., `pnpm knip`) if configured.
Report: PASS or FAIL with unused exports/dependencies found.
Suggest removals for each finding.

Task 5: Unit Tests

Run unit tests (e.g., `pnpm test:unit` or per AGENTS.md) in the target repository.
Report: PASS with test count, or FAIL with failing test names and errors.
For failures, include the assertion that failed.

Task 6: Style Lint (only if CSS changes exist)

Run style linting (e.g., `pnpm stylelint`) if configured.
Report: PASS or FAIL with issues found.

3. Collect and Report Results

Wait for all subagents, then compile report:

# Quality Gates Report

## Summary

| Gate       | Status            | Issues  |
| ---------- | ----------------- | ------- |
| Lint       | ✅ PASS / ❌ FAIL | {count} |
| Format     | ✅ PASS / ❌ FAIL | {count} |
| Typecheck  | ✅ PASS / ❌ FAIL | {count} |
| Knip       | ✅ PASS / ❌ FAIL | {count} |
| Unit Tests | ✅ PASS / ❌ FAIL | {count} |
| Stylelint  | ✅ PASS / ❌ FAIL | {count} |

## Overall: {PASS / FAIL}

## Issues to Fix

### Lint Errors

{file:line - error message - suggested fix}

### Type Errors

{file:line - error message - suggested fix}

### Knip Findings

{unused export/dependency - removal suggestion}

### Test Failures

{test name - assertion that failed}

4. Handle Results

All gates pass:

• Update status.json: set status to "review"
• Prompt to continue to code review phase

Any gates fail:

• Present issues clearly
• Ask: "Fix these issues automatically?" (Y/N)
• If Y, proceed to auto-fix flow
• If N, user fixes manually

5. Auto-Fix Flow

For auto-fixable issues, run directly:

For non-auto-fixable issues, dispatch fix subagents:

Type Error Fix Subagent:

Fix the following type error:
File: {file}
Line: {line}
Error: {error message}
Context: {surrounding code}
Apply the fix and verify with the typecheck command.

Test Failure Investigation Subagent:

Investigate failing test:
Test: {test name}
File: {test file}
Assertion: {failed assertion}
Determine root cause and fix. Re-run tests to verify.

6. Re-run Failing Gates

After fixes:

• Re-run only the gates that failed
• Loop until all pass or user stops

Quality Gate Commands

Key Principles

1. Run in parallel: All checks run simultaneously for speed
2. Clean context: Only summary returns to main agent
3. Auto-fix safely: Format issues fix automatically without asking
4. Show before applying: For complex fixes (type errors), show fix before applying
5. Track passed gates: Don't re-run gates that already passed