spec-harness-learn

To stop AI coding assistants from repeating errors, you must capture correction feedback as permanent, dated rules in an AGENTS.md ratchet system. This ensures the AI loads and enforces these tightening rules at startup for all future development sessions.

A ratchet system for AI development is an append-only rule set where corrections are codified as dated rules that only ever tighten. It ensures coding patterns remain consistent over time by preventing previously enforced rules from being removed or relaxed.

You convert verifier failure root causes into enforced coding rules by running a learning loop that distills the failure into a single imperative rule, classifies it, and injects it as a dated rule into your AGENTS.md startup load path.

Yes, spec-harness-learn works directly with verifier-driven development cycles by capturing verifier failure root causes and converting them into permanent, dated rules that are automatically loaded and enforced in future AI-assisted development sessions.

You maintain an append-only tightening rule set by capturing AI correction feedback and injecting it as a dated rule into your AGENTS.md file. This ratchet approach ensures rules are appended or sharpened, never removed, eliminating recurring errors.

Yes, you can automate capturing AI correction feedback through a 6-step learning loop that distills the feedback into a single imperative rule, classifies it to the correct location, injects it as a dated rule, and verifies it loads at startup.