try-learn-skill

Workflow analysis identifies reusable procedures by reviewing planning docs, task files, and git diffs from finished work. It applies high-threshold filtering to reject generic implementation steps, keeping only non-obvious procedures with durable value for future projects.

Turn completed work into a learned skill when post-implementation analysis of artifacts reveals a non-obvious, repeatable procedure. If the feature only involved normal engineering work without durable patterns, it should not be promoted to a learned skill.

Track learned skills via a ledger that records first sightings of patterns as candidates. Repeated patterns are promoted into learned skills only after passing collision checks against existing static skills and receiving explicit human approval.

The best way to filter generic implementation steps is applying high-threshold criteria during workflow analysis. Reject obvious one-off work and retain only non-obvious procedures that demonstrate durable, repeatable value for future engineering projects.

Yes, promoting a reusable workflow into a learned skill requires human approval. The ledger-first learning process records candidates, checks for collisions with existing skills, and mandates explicit human approval before creating any new learned skill.