菜单
Record and analyze recommendation sessions and feedback to drive the self-learning loop
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Checkpoint episodes and create handoff packs for multi-agent session continuity
Configure, test, benchmark, and manage embedding providers for semantic search and vector operations
Create and traverse typed relationships between episodes for dependency tracking and knowledge graphs
Manage episode tags for organization, filtering, and retrieval across the memory system
Generate task playbooks and explain patterns using learned strategies from episodic memory
Use and troubleshoot the Memory MCP server for episodic memory retrieval and pattern analysis. Use when working with MCP server tools, validating the MCP implementation, or debugging MCP server issues.
| name | recommendation-feedback |
| description | Record and analyze recommendation sessions and feedback to drive 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
| 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 |
| 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 |
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"
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
| Value | Meaning |
|---|---|
success | Task completed successfully using recommendations |
partial | Some recommendations helped, others didn't |
failure | Recommendations were not useful |
do-memory-cli feedback stats
Returns: total sessions, average rating, most-applied patterns, patterns with highest success correlation.
Feedback data directly influences rank_patterns() in future queries: