session-logger（会话记录器）

Session Logger

A skill for automatically saving conversation history to persistent session log files.

When This Skill Activates

This skill activates when you:

• Say "保存对话信息" or "保存对话"
• Say "记录会话内容" or "保存session"
• Say "save session" or "save conversation"
• Ask to save the current conversation

Session File Location

All sessions are saved to: sessions/YYYY-MM-DD-{topic}.md

What Gets Logged

For each session, log:

1. Metadata

- Date and duration - Context/working directory - Main topic

1. Summary

- What was accomplished - Key decisions made - Files created/modified

1. Actions Taken

- Checklist of completed tasks - Pending follow-ups

1. Technical Notes

- Important code snippets - Commands used - Solutions found

1. Open Questions

- Issues to revisit - Follow-up tasks

Session Template

# Session: {Topic}

**Date**: {YYYY-MM-DD}
**Duration**: {approximate}
**Context**: {project/directory}

## Summary

{What was accomplished in this session}

## Key Decisions

1. {Decision 1}
2. {Decision 2}

## Actions Taken

- [x] {Completed action 1}
- [x] {Completed action 2}
- [ ] {Pending action 3}

## Technical Notes

{Important technical details}

## Open Questions / Follow-ups

- {Question 1}
- {Question 2}

## Related Files

- `{file-path}` - {what changed}

How to Use

Option 1: Automatic Logging

Simply say:

"保存对话信息"

The skill will:

1. Review the conversation history
2. Extract key information
3. Create/update the session file

Option 2: With Topic

Specify the session topic:

"保存对话，主题是 skill-router 创建"

Option 3: Manual Prompt

If auto-extraction misses something, provide details:

"保存对话，重点是：1) 创建了 skill-router，2) 修复了 front matter"

File Naming

Session Log Structure

sessions/
├── README.md                      # This file
├── 2025-01-11-skill-router.md     # Session about skill-router
├── 2025-01-11-prd-planner.md      # Session about PRD planner
└── 2025-01-12-refactoring.md      # Session about refactoring

Privacy Note

Session logs are stored in sessions/ which is in .gitignore.

• Logs are NOT committed to git
• Logs contain your actual conversation
• Safe to include sensitive information

Quick Reference

Best Practices

1. Save at key milestones: After completing a feature, fixing a bug, etc.
2. Be specific with topics: Helps when searching later
3. Include code snippets: Save important solutions
4. Track decisions: Why did you choose X over Y?
5. List pending items: What to do next time

Rich Content Extraction (for Self-Improving Agent)

When triggered by other skills via hooks, session-logger extracts structured data for learning:

Skill Context Capture

When a skill completes, capture:

## Skill Execution Context

**Skill**: {skill-name}
**Trigger**: {user-invoked | hook-triggered | auto-triggered}
**Status**: {completed | error | partial}
**Duration**: {approximate time}

### Input Context
- User request: {original request}
- Files involved: {list of files}
- Codebase patterns detected: {patterns}

### Output Summary
- Actions taken: {list}
- Files modified: {list with changes}
- Decisions made: {key decisions}

### Learning Signals
- What worked well: {successes}
- What could improve: {areas for improvement}
- Patterns discovered: {new patterns}
- Errors encountered: {errors and resolutions}

Error Context Capture

When a skill encounters errors:

## Error Context

**Error Type**: {type}
**Error Message**: {message}
**Stack Trace**: {if available}

### Resolution Attempted
- Approach: {what was tried}
- Result: {success/failure}
- Root cause: {if identified}

### Prevention Notes
- How to avoid: {prevention strategy}
- Related patterns: {similar issues}

Pattern Extraction

Extract reusable patterns for the self-improving-agent:

## Extracted Patterns

### Code Patterns
- Pattern name: {name}
- Context: {when to use}
- Example: {code snippet}

### Workflow Patterns
- Trigger: {what initiates}
- Steps: {sequence}
- Outcome: {expected result}

### Anti-Patterns
- Pattern: {what to avoid}
- Why: {reason}
- Alternative: {better approach}

Structured Data Format

For machine-readable extraction, use YAML front matter in session logs:

---
session_type: skill_execution
skill_name: code-reviewer
trigger_source: hook
status: completed
files_modified:
  - path: src/utils.ts
    changes: refactored error handling
patterns_learned:
  - name: error-boundary-pattern
    category: error-handling
    confidence: high
errors_encountered: []
learning_signals:
  successes:
    - "Identified code smell in utils.ts"
  improvements:
    - "Could have suggested more specific refactoring"
---

Integration with Self-Improving Agent

When triggered by self-improving-agent:

1. Extract episodic memory: Capture the full context of what happened
2. Identify semantic patterns: Tag reusable knowledge
3. Update working memory: Note immediate follow-ups needed
4. Signal completion: Write trigger file if skill chaining is needed

Auto-Trigger Behavior

When invoked via hooks with mode: auto:

• Silently create/update session log
• Extract structured data without user interaction
• Append to existing session if same day/topic
• Create new session if context differs significantly