Verification Loop

Verification Loop

Run all quality gates before committing or creating a PR.

When to Use

• Before any git commit
• Before creating a pull request
• After completing implementation (before marking done)
• When asked to "verify" or "check everything"

Procedure

Step 1: Detect Project Type

Identify available tools from project files:

• package.json: npm/yarn scripts (lint, typecheck, test)
• pyproject.toml / setup.py: Python tooling (ruff, mypy, pytest)
• go.mod: Go tooling (golangci-lint, go vet, go test)
• Cargo.toml: Rust tooling (clippy, cargo test)

Step 2: Run Quality Gates

Execute in order (stop on CRITICAL):

| Gate | Command (detect per project) | Severity | |------|------------------------------|----------| | Lint | eslint, ruff, golangci-lint | HIGH | | Type Check | tsc --noEmit, mypy, go vet | HIGH | | Tests | jest, pytest, go test | CRITICAL | | Security | npm audit, pip-audit, cargo audit | CRITICAL | | Coverage | jest --coverage, pytest --cov | HIGH |

Step 3: Report Results

Verification: {project name}

  Lint:       PASS | FAIL ({count} issues)
  Types:      PASS | FAIL ({count} errors)
  Tests:      PASS | FAIL ({pass}/{total})
  Security:   PASS | FAIL ({count} vulnerabilities)
  Coverage:   {percent}% (target: 80%)

Verdict: READY TO COMMIT | BLOCKED ({reason})

Step 4: Actionable Fixes

For each failure, provide:

{gate} FAIL:
  {file}:{line} - {error message}
  Fix: {specific suggestion}

Blocking Rules

• CRITICAL failures block commit (tests failing, known vulnerabilities)
• HIGH failures: warn but allow commit with acknowledgment
• MEDIUM/LOW: report only, do not block

Output Format

Verification: lead-scoring-simulator

  Lint:       PASS
  Types:      PASS
  Tests:      PASS (24/24)
  Security:   PASS
  Coverage:   87% (target: 80%)

Verdict: READY TO COMMIT

Rules

• Always run tests with verbose output
• Kill any hung processes after 120s timeout
• If a gate tool is not installed, skip it with a warning
• Never auto-fix lint issues without showing what changed
• Clean up test artifacts after run

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: mshadmanrahman
• Source: mshadmanrahman/pm-pilot
• License: MIT
• Homepage: https://github.com/mshadmanrahman/pm-pilot#readme

Install and usage instructions live in the source repository linked above.