| name | recommendation-feedback |
| description | Record and analyze recommendation sessions and feedback to drive the self-learning loop |
Recommendation Feedback
When to Use
- Recording which patterns/playbooks were recommended to an agent
- Capturing feedback on whether recommendations were useful
- Analyzing recommendation effectiveness statistics
- Closing the self-learning feedback loop
The Self-Learning Loop
1. Agent starts task → query_memory / recommend_patterns
2. System recommends patterns/playbooks → record_recommendation_session
3. Agent completes task → record_recommendation_feedback
4. System adjusts future recommendations based on outcomes
CLI Commands
| Command | Purpose |
|---|
do-memory-cli feedback record-session -e <ID> -p <patterns> -P <playbooks> | Record what was recommended |
do-memory-cli feedback record-feedback -s <SESSION> -a <applied> -o <outcome> -m <msg> | Record what worked |
do-memory-cli feedback stats | View recommendation statistics |
MCP Tools
| Tool | Parameters | Purpose |
|---|
record_recommendation_session | episode_id, patterns, playbooks | Log recommended items |
record_recommendation_feedback | session_id, applied, consulted, outcome, rating | Log what was used |
get_recommendation_stats | - | Aggregate effectiveness stats |
Recording a Session
When patterns are recommended to an agent:
do-memory-cli feedback record-session \
--episode-id "abc-123" \
--patterns "pat-1,pat-2,pat-3" \
--playbooks "pb-1"
Recording Feedback
After the episode completes:
do-memory-cli feedback record-feedback \
--session "session-uuid" \
--applied "pat-1,pat-3" \
--consulted "ep-old-1" \
--outcome "success" \
--message "Pattern 1 was directly applicable, pat-2 was irrelevant" \
--rating 0.8
Outcome Values
| Value | Meaning |
|---|
success | Task completed successfully using recommendations |
partial | Some recommendations helped, others didn't |
failure | Recommendations were not useful |
Viewing Stats
do-memory-cli feedback stats
Returns: total sessions, average rating, most-applied patterns, patterns with highest success correlation.
Integration with Pattern Ranking
Feedback data directly influences rank_patterns() in future queries:
- Patterns with high apply+success rates get boosted
- Patterns frequently recommended but never applied get demoted
- This is the core mechanism that makes the memory system self-improving
