Extract Learnings
Capture reusable knowledge from completed work into .feature-radar/specs/.
Deep Read
Behavioral Directives
Workflow
- Identify the source — ask the user what was just completed (feature, bug fix, refactor, investigation)
- Analyze the work — review recent commits, changed files, and implementation decisions
- Extract knowledge — identify what's reusable:
- Patterns: recurring solutions worth replicating (e.g., "three-tier config merge") - Decisions: architectural choices with rationale (e.g., "YAML over JSON because...") - Pitfalls: mistakes or dead ends others should avoid - Techniques: implementation approaches that worked well
Before writing to specs/, classify each piece of knowledge into exactly one category:
- Pattern: recurring solution worth replicating
- Decision: architectural choice with rationale
- Pitfall: mistake or dead end to avoid
- Technique: implementation approach that worked well
State the classification explicitly in your output.
- Write to specs — create or append to
.feature-radar/specs/{topic}.md - Checkpoint — State what was written and ask: "I've written to
specs/{topic}.md({classification type}). Does this look correct, or should I adjust anything?" Wait for user confirmation before proceeding. - Update base.md — increment the specs count in Tracking Summary
File Format
Use the format defined in ../feature-radar/references/SPEC.md § 3.4 (specs/{topic}.md).
Guidelines
- One topic per file. If the learning spans multiple topics, create multiple files.
- Name files by the pattern, not by the feature that produced it.
- Good: yaml-config-merge.md, symlink-vs-copy-tradeoffs.md - Bad: audit-feature-learnings.md, v2-refactor-notes.md
- Append to existing files when the new learning extends a known topic.
- Keep it concise — future readers need the insight, not the full story.
Example Output
→ Created specs/symlink-vs-copy-tradeoffs.md (Decision)
→ Updated base.md: specs 2 → 3Completion Summary
Follow the template in ../feature-radar/references/DIRECTIVES.md, with skill name "Learn Complete".