verification-loop_skill

Overview

This skill implements a comprehensive 6-check verification framework to validate implementation quality across build, types, lint, tests, security, and diff review. It runs deterministically, detects project type automatically, and acts as a pre-merge or implement-phase gate. The goal is to catch issues early and ensure code meets quality gates before proceeding.

How this skill works

The verification loop auto-detects the project type (Node/TypeScript, Python, Go, Rust, or mixed) and selects appropriate commands for each check. It runs checks in a fail-fast order: build, types, lint, tests, security, and diff review, parsing outputs and producing structured pass/fail results. For failures it extracts precise error context, optionally spawns targeted fix subagents, retries up to a configurable limit, and re-runs checks to confirm fixes.

When to use it

• Automatically as implement-phase Step 2 (exit condition verification)
• Before integration testing or merging feature branches
• Validating changes before committing or CI runs
• After refactors to ensure no regressions
• Not for read-only analysis or single-file exploratory checks

Best practices

• Run the loop locally before opening a PR to reduce CI friction
• Prefer project-local tooling (poetry/pdm, npm/yarn/pnpm) to match CI behavior
• Allow the loop to auto-fix mechanical lint issues before failing
• Configure sensible coverage and lint thresholds in project config
• Use the diff check to prevent accidental or out-of-scope changes

Example use cases

• Pre-merge verification to block broken or insecure changes
• CI implement-phase exit condition to gate progress to integration testing
• Local developer workflow to catch build, type, or lint failures early
• Post-refactor validation to ensure types and tests still pass
• Security pre-release scan to detect secrets or vulnerable dependencies