apply-learnings

To turn user feedback into permanent skill improvements, this process reads pending corrections from a learnings queue JSONL file, generalizes them into reusable rules, and writes them directly into target skill SKILL.md files. User confirmation is required before any generalized rule is applied.

Rule generalization works by using the Reflect-Abstract-Generalize-Write pattern to convert specific, raw user corrections into absolute, reusable directives. These generalized rules are then added to the skill's active learnings section to ensure compliance in all future data modeling executions.

The best way to stop repeated execution errors is to systematically process user corrections through a feedback integration workflow. By grouping pending corrections by associated skill and applying approved generalized rules, future executions automatically avoid the same mistakes.

Yes, you can review and edit generalized rules before application. A user approval workflow presents proposed rules via an interactive prompt, allowing you to confirm, reject, or edit them to ensure the updates align with your expectations before they modify the target skill files.

Yes, managing learnings requires read and write access to the learnings queue JSONL file and target SKILL.md files. This access allows the system to process pending correction groups and permanently update the active learnings sections across all skills in the data modeler plugin ecosystem.

Your data modeling skill repeats mistakes because raw corrections are not being systematically integrated into its workflow. Without applying learnings to generalize your feedback into permanent rules and updating the skill files, past corrections are lost and execution errors persist.