What problem does it solve?
Developers and AI agents often receive corrective feedback during implementation but lack a structured way to decide whether that feedback should become a permanent rule, a test, a design note, or be discarded. This Skill prevents skills and instructions from becoming append-only knowledge dumps by enforcing classification and compaction rules.
Core Features & Use Cases
- Feedback Classification: Routes corrections to the correct destination: regression tests, design documents, instruction files, learning inboxes, or skills.
- Learning Inbox Management: Creates and maintains
.github/learnings/<area>.md files with strict size and entry limits, including compaction and garbage collection rules.
- Use Case: When a user rejects a specific implementation pattern during code review, use this Skill to determine whether the correction is a one-off preference, a regression test case, or a stable architectural invariant that belongs in a design document.
Quick Start
Apply the feedback-learning skill to classify this user correction and record it in the appropriate repository artifact.
