spec-harness-verify

To verify code meets project goals beyond automated tests, use an independent clean-context agent to check implementation against explicit requirements. This literal, inference-free verification evaluates whether end-state requirements from goal.md are fully met and returns clear PASS or FAIL results.

Using the same AI agent to verify code leads to biased checks because the agent retains context from the code-writing process, causing inaccurate evaluations that miss unmet requirements. A separate, clean-context verifier eliminates this bias by assessing implementation without assumptions from the development phase.

Literal requirement checking in spec-driven development assesses every end-state requirement exactly as written, with no inference or assumption of unstated intent. It evaluates implementation against goal.md to ensure the code explicitly satisfies documented project goals.

To integrate code verification failures into a learning workflow, feed all FAIL results and correction causes into a spec-harness-learn process. This creates dated, sharpened improvement rules that capture failure causes and enhance future software development workflows.

You can use independent verification for software development workflows where passing automated tests does not guarantee end-state requirements are fully met. It is suitable for any project utilizing explicit goal documents to validate implementation through clean-context agent checking.