feedback

Compound Knowledge Feedback Manager

Persist feedback about knowledge capture and retrieval across sessions. Stored feedback is automatically loaded when compound-capture or compound-retrieve runs, ensuring corrections are never repeated.

Actions

Save Feedback

When the user provides feedback on capture or retrieval behavior, save it:

echo '{"category": "<category>", "feedback": "<what the user said>", "context": "<optional context>"}' | \
  python ${PLUGIN_ROOT}/scripts/feedback_manager.py compound-knowledge save-feedback

Categories: capture_format, categorization, triviality_threshold, cross_references, retrieval_depth, principle_style, general

Examples:

• "Don't capture trivial one-line fixes" → {"category": "triviality_threshold", "feedback": "Raise the bar — only capture solutions that required multiple investigation steps"}
• "Always include code examples" → {"category": "capture_format", "feedback": "Include inline code examples in the solution body, not just descriptions"}
• "Stop putting patterns in debugging/" → {"category": "categorization", "feedback": "Design patterns belong in patterns/, even if discovered during debugging"}
• "Show more results when searching" → {"category": "retrieval_depth", "feedback": "Return top 5-7 results instead of top 3 when searching solutions"}
• "Principles should be shorter" → {"category": "principle_style", "feedback": "Keep principle statements to one sentence maximum"}

Show Feedback

Display all stored feedback:

python ${PLUGIN_ROOT}/scripts/feedback_manager.py compound-knowledge show-feedback

Present as a readable list grouped by category.

Clear Feedback

Clear all feedback or feedback for a specific category:

# Clear all
python ${PLUGIN_ROOT}/scripts/feedback_manager.py compound-knowledge clear-feedback

# Clear only categorization feedback
python ${PLUGIN_ROOT}/scripts/feedback_manager.py compound-knowledge clear-feedback categorization

Consolidate Feedback

Graduate stable feedback into the actual SKILL.md files, making corrections permanent. This is a Claude-driven operation — no script needed.

When to consolidate: When the user says "update the plugin based on feedback", "consolidate feedback", "bake in my preferences", or "graduate feedback into the skill".

Process:

1. Load all stored feedback:

python ${PLUGIN_ROOT}/scripts/feedback_manager.py compound-knowledge show-feedback

2. Read the target SKILL.md files that the feedback applies to:

   • ${PLUGIN_ROOT}/skills/compound-capture/SKILL.md
   • ${PLUGIN_ROOT}/skills/compound-retrieve/SKILL.md

3. For each feedback entry, determine if it should be consolidated:

   • Consolidate: Universal preferences, repeated corrections, style rules that always apply
   • Keep as runtime feedback: Situational preferences, context-dependent corrections, temporary focus areas

4. Present a consolidation plan to the user:

   ## Consolidation Plan
   
   **Will bake into SKILL.md** (permanent):
   - [feedback] → edit [file]: [what will change]
   
   **Will keep as runtime feedback** (situational):
   - [feedback] → reason: [why it stays runtime]
   
   Proceed?
   

5. On approval:

   • Edit the target SKILL.md files using the Edit tool
   • Clear only the graduated feedback entries:

   python ${PLUGIN_ROOT}/scripts/feedback_manager.py compound-knowledge clear-feedback <category>
   

   • Keep non-graduated entries untouched

6. Report what changed and what remains as runtime feedback.

How Feedback is Used

When compound-capture runs, it loads all feedback entries and applies them:

• capture_format feedback → adjusts the structure and detail level of solution files
• categorization feedback → guides which directory a solution is filed into
• triviality_threshold feedback → calibrates what gets captured vs skipped
• cross_references feedback → adjusts how aggressively related solutions are linked
• principle_style feedback → shapes how engineering principles are worded
• general feedback → applied to all capture and retrieval operations

When compound-retrieve runs, it loads retrieval-related feedback:

• retrieval_depth feedback → adjusts how many results are surfaced
• general feedback → applied to result formatting and presentation

This ensures the user never has to repeat the same correction twice.