Learning Playbook

Treat observations as evidence until an owner deliberately promotes them. This skill reviews promotion candidates; it does not capture them or make them govern future work.

Remain read-only by default. A request to review evidence does not authorize edits to repositories, knowledge bases, skills, policy, or agent configuration.

Place in the learning workflow

Use /learning-playbook to assess whether learning from saved findings and real project work should become a reusable Playbook or another durable owner. Its first output is a promotion verdict with the proposed destination, concrete change, evidence and applicability boundary. A lesson can remain local; the name does not require a Playbook for every finding.

Review stays read-only until the user approves the proposed write or an explicit applicable grant covers it. After approval, apply the accepted change at its named owner using that owner's promotion and completion rules. Candidate status and required behavioral proof remain explicit; approval to record a candidate does not establish general effectiveness. Use /learning-save for an authorized KB record of the accepted verdict and provenance. Use /learning-loop to inspect later application and return evidence to the owner.

Establish The Evidence Boundary

Before assessing a candidate, identify: