spec-harness-learn

What problem does it solve?

This Skill eliminates the risk of repeating the same AI coding errors after corrections or verifier failures by capturing lessons as permanent, dated rules that are automatically enforced in all future development sessions.

Core Features & Use Cases

• 6-Step Learning Loop: Captures feedback, distills it to a single imperative rule, classifies it to the correct location in the project's rule set, injects it as a dated rule, optionally generates a human-readable lesson, and verifies the rule is loaded at startup.
• Ratchet Enforcement: Ensures the project's AGENTS.md rule set only ever tightens (rules are appended or sharpened, never removed) to maintain consistent coding patterns over time.
• Use Case: If you correct an AI for generating incorrect Redux Toolkit slice structure, this Skill will add a dated rule to AGENTS.md that enforces the correct structure for all future code generation, eliminating the need to re-explain the pattern every session.

Quick Start

Use the spec-harness-learn skill to capture the correction you just gave for the incorrect API endpoint implementation as a permanent enforced rule.