run-and-fix

name	run-and-fix
description	Runs project quality gates in sequence, auto-fixes failures, retries up to N times per step before escalating. Adapts to any runtime via project.yaml.

Run and Fix

Executes quality gates sequentially, fixing failures along the way.

Step 1: Load project config

Read project.yaml (in the project's config directory, default .ai/) to get:

• quality_gates.commands (format, lint, typecheck, test, audit)
• quality_gates.retry_max (default: 3)
• quality_gates.coverage_threshold

If no project.yaml exists, detect from project root files:

• deno.json / deno.jsonc -> deno fmt / deno lint / deno task check / deno task test
• package.json -> npm run format / npm run lint / tsc --noEmit / npm test
• go.mod -> gofmt / golangci-lint run / (skip typecheck) / go test ./...
• Cargo.toml -> cargo fmt / cargo clippy / (skip typecheck) / cargo test

Step 2: Execute gates in order

Run each step. On failure, analyze the error, apply a minimal fix, re-run. Move to the next step only when the current one passes.

Gate sequence

1. Format - Run format command. Auto-fixes formatting. Rarely fails.
2. Lint - Run lint command. On failure: read error, apply fix, re-run.
3. Type check - Run typecheck command (skip if null). On failure: fix types, re-run.
4. Tests - Run test command. On failure: fix code or test, re-run. Do NOT weaken assertions.
5. Audit - Run audit command (skip if null). On failure: report to user, do NOT auto-fix.

Iteration Protocol

for each step:
  attempt = 0
  while step fails AND attempt < retry_max:
    analyze error output
    apply minimal targeted fix
    re-run step
    attempt++
  if step still fails:
    STOP and report to user with:
      - which step failed
      - error output
      - what was tried
      - suggested manual fix

Step 3: Coverage verification (optional)

If quality_gates.commands.coverage is defined:

1. Run coverage command.
2. Verify coverage_threshold% minimum line coverage on modified files.
3. If below threshold: identify uncovered branches, report to user. Do NOT generate weak tests.
4. BLOCK until threshold is met.

Step 4: Git hygiene (optional)

If running as a final pre-commit check:

git diff --stat main...HEAD

Check for:

• Unintended file changes (lock files, unrelated files)
• Files that should be in .gitignore (coverage/, dist/, node_modules/)
• Large binary files or generated content

Report findings but do not auto-fix.

Output Format

## Quality Gate Results

| Step       | Status | Attempts | Notes                    |
|------------|--------|----------|--------------------------|
| Format     | PASS   | 1        | Auto-fixed 3 files       |
| Lint       | PASS   | 2        | Fixed unused import      |
| Type check | PASS   | 1        | -                        |
| Tests      | PASS   | 1        | 42 tests, all passing    |
| Audit      | PASS   | 1        | No vulnerabilities found |
| Coverage   | PASS   | -        | 85% (threshold: 80%)     |
| Git hygiene| PASS   | -        | No unintended changes    |

All gates passed.

Rules

• ALWAYS run format first (it fixes issues other steps would flag).
• NEVER weaken a test assertion to make it pass.
• NEVER skip a step (except null commands).
• NEVER auto-fix security audit issues without user approval.
• After fixing lint/typecheck errors, re-run format to ensure consistency.
• Follow project conventions from CLAUDE.md for any code changes.