pre-pr-check

Pre-PR Check

When to invoke

BEFORE opening a PR, BEFORE merging to main, BEFORE pushing to a tracked branch. Invoked by supervibe:finishing-a-development-branch and supervibe:requesting-code-review.

Expert Operating Standard

Follow docs/references/skill-expert-operating-standard.md: start from source of truth, preserve retrieval evidence, apply scope safety, use real producers with runtime receipts for durable delegated outputs, verify before completion claims, and keep confidence below gate when evidence is partial.

Step 0 — Read source of truth (required)

1. Read project's check command(s) from the active host instruction file or package.json/composer.json/Cargo.toml/Makefile
2. Identify stack (decides which checks apply)
3. Note any project-specific gates (e.g., bundle size budget, coverage threshold)

Decision tree

What stack determines which checks?
├─ Node.js → tsc --noEmit + npm test + npm run lint + npm audit + bundle-size (if applicable); never auto-apply `npm audit fix --force` when it downgrades a framework major/minor line
├─ Python → mypy + pytest + ruff + pip-audit + safety check
├─ PHP → phpstan + pest/phpunit + pint + composer audit
├─ Go → go vet + go test + golangci-lint + govulncheck
├─ Rust → cargo check + cargo test + cargo clippy + cargo audit
└─ Mixed → all above for each detected stack

Procedure

1. Run stack-appropriate checks sequentially (parallel risks output mixing)
2. Capture each output verbatim (no paraphrasing)
3. Decide pass/fail per check:
   • Typecheck: 0 errors
   • Tests: 0 failures, 0 regressions
   • Lint: 0 errors (warnings may be allowed per project)
   • Audit: 0 high+critical vulns
4. If ANY fail → STOP, return failure with output; caller must fix
5. If ALL pass → emit evidence bundle
6. Score — supervibe:confidence-scoring artifact-type=agent-output; ≥9 required

Output contract

Returns evidence bundle:

✓ Typecheck: <output snippet> (exit 0)
✓ Tests: <count passed/failed> (<output>)
✓ Lint: <output> (exit 0)
✓ Audit: <count vulns by severity>
✓ Bundle size: <delta vs baseline> (if applicable)

Verdict: READY / NOT-READY

Guard rails

• DO NOT: skip a check because "it usually passes"
• DO NOT: paraphrase output ("looks good" without showing it)
• DO NOT: declare ready with non-zero exit code anywhere
• DO NOT: ignore audit findings (must fix or document override)
• ALWAYS: capture output verbatim
• ALWAYS: include exit codes

Verification

• Every required check ran and exit code captured
• Output is verbatim, not summarized
• Verdict matches results

Related

• supervibe:finishing-a-development-branch — invokes this first
• supervibe:requesting-code-review — invokes this before reviewer
• supervibe:verification — per-claim verification (this is per-PR)