agent-self-reflection（Agent 人自我反思）

Self-Reflection Skill

Reflect on recent sessions and extract actionable insights. Runs hourly via cron.

Step 1: Gather Recent Sessions

# List sessions active in the last 2 hours
openclaw sessions --active 120 --json

Parse the output to get session keys and IDs. Skip subagent sessions (they're task workers, not interesting for reflection). Focus on:

• Telegram group/topic sessions (real user interactions)
• Direct sessions (1:1 with Brenner)
• Cron-triggered sessions (how did automated tasks go?)

Step 2: Read Session History

For each interesting session from Step 1, read the JSONL transcript:

# Read the last ~50 lines of each session file (keep it bounded!)
tail -50 ~/.openclaw/agents/main/sessions/<sessionId>.jsonl

⚠️ CRITICAL: Never load full session files. Use tail -50 or Read with offset/limit. Sessions can be 100k+ tokens.

Parse the JSONL to understand what happened. Look for:

• type: "user" or type: "human" — what was asked
• type: "assistant" — what you responded
• type: "tool_use" / type: "tool_result" — what tools were called and results
• Error patterns, retries, confusion

Step 3: Analyze & Extract Insights

For each session, ask yourself:

What went well?

• Tasks completed smoothly on first try
• Good tool usage patterns worth reinforcing
• Efficient approaches to remember

What went wrong?

• Errors, retries, wrong approaches
• Misunderstandings of user intent
• Tools that didn't work as expected
• Context that was missing

Lessons learned?

• "Next time, do X instead of Y"
• "Remember that Z works this way"
• "Tool A needs parameter B or it fails"
• "When user says X, they usually mean Y"

Quality bar: Each insight must be:

• Specific — not "be more careful" but "check if file exists before editing"
• Actionable — something future-you can directly apply
• Non-obvious — skip things any competent agent would know
• New — don't repeat insights already captured

Step 4: Route Insights to the Right Files

Each insight belongs somewhere specific. Route them:

→ AGENTS.md

• Process improvements (how to handle sessions, memory, etc.)
• New conventions or workflow rules
• Safety lessons

→ TOOLS.md

• Tool-specific gotchas ("gog needs --json flag for parsing")
• Environment details (paths, configs, quirks)
• New tool patterns discovered

→ memory/YYYY-MM-DD.md (today's date)

• Session-specific context ("Brenner asked about X project")
• Temporary facts that matter today but not forever
• What happened today (events, decisions, requests)

→ memory/about-user.md

• New preferences discovered
• Communication style observations
• Project/interest updates

→ skills/<skill-name>/SKILL.md

• Improvements to specific skill instructions
• Bug fixes in skill workflows
• New parameters or approaches for a skill

→ MEMORY.md

• Updates to the memory index if new memory files are created

Step 5: Write the Insights

For each insight, append or edit the appropriate file. Use the Edit tool for surgical changes to existing content. Use append (write to end) for daily memory files.

Format for daily memory files:

## Self-Reflection — HH:MM ET

### Insights
- [source: session-key] Lesson learned here
- [source: session-key] Another insight

### Tool Notes
- Discovered: tool X needs Y configuration

### User Context
- Brenner mentioned interest in Z

Step 6: Summary

After writing all insights, produce a brief summary of what you reflected on and what you wrote. This is your output — keep it to 2-4 sentences max.

If there's nothing interesting to reflect on (quiet period, only heartbeats), just say so. Don't manufacture insights.

Quality Checklist

Before writing any insight:

• Is this actually new? (Check existing files first)
• Is this specific and actionable?
• Am I routing it to the right file?
• Am I keeping daily memory files concise (not dumping full transcripts)?
• Did I respect the token budget (no huge file reads)?

Anti-Patterns (Don't Do These)

• ❌ Don't summarize every session — only extract *lessons*
• ❌ Don't read full JSONL files — tail/limit only
• ❌ Don't write vague insights ("improve response quality")
• ❌ Don't duplicate existing knowledge
• ❌ Don't create new files when appending to existing ones works
• ❌ Don't reflect on your own reflection sessions (skip cron:self-reflection sessions)