| name | reflect |
| description | Review chat session for mistakes, friction, and improvement opportunities; analyze and improve skills used |
| license | MIT |
| compatibility | pi |
| metadata | {"audience":"all-users","workflow":"quality-improvement"} |
Success Criteria
Before presenting output, verify:
- ✅ Chat review covers mistakes, friction points, and unclear outputs
- ✅ Concrete, actionable improvements proposed (not vague suggestions)
- ✅ Skills used in chat are analyzed for self-checks, conciseness, and quality
- ✅ Pattern detection identifies repeated tasks that could become skills
- ✅ Clear questions asked about what to remember for future chats
Iterate up to 5 times if criteria not met.
What I Do
When the user types "reflect", I perform a comprehensive review of the current chat session:
1. Chat Session Review
- Analyze what we've accomplished
- Identify mistakes, errors, or incorrect assumptions
- Note friction points (repeated clarifications, confusion, corrections)
- Flag unclear or incomplete outputs
- Detect frustration signals from user
2. Improvement Proposals
- Generate concrete, actionable improvements (not vague suggestions)
- Prioritize high-impact changes
- Present as a short numbered list
- Ask which improvements to remember for future sessions
3. Skill Quality Analysis
For each skill used during this chat:
a) Self-Check Validation:
- Check if skill has success criteria at the top
- Check if skill has verification instruction at bottom
- If missing: propose adding both with specific criteria for that skill
- Ensure verification includes "iterate up to 5 times if not met"
b) Token Efficiency:
- Identify verbose sections that could be more concise
- Remove redundant instructions
- Suggest structural improvements
- Preserve clarity and functionality
c) General Improvements:
- Check for outdated patterns
- Identify missing edge cases
- Suggest better organization
- Propose enhanced instructions
4. Pattern Detection
- Monitor for tasks I'm asked to do repeatedly
- Suggest creating reusable skills for common patterns
- Ask for confirmation before creating new skills
- Provide skill name and description suggestions
5. Proactive Suggestions
Suggest typing "reflect" when:
- User corrects me multiple times on same topic
- User clarifies something twice or more
- User shows signs of frustration
- Workflow feels inefficient or repetitive
When to Use Me
Trigger explicitly when user types:
reflect
review session
how are we doing?
Suggest proactively when:
- User corrects the same mistake 2+ times
- User re-explains requirements
- Multiple back-and-forth exchanges on unclear point
- User expresses frustration (e.g., "no, that's not what I meant")
- Session feels inefficient
Output Format
Session Review
Accomplishments:
Mistakes/Issues:
- [Specific mistake with context]
- [Friction point with example]
Unclear Outputs:
- [What was unclear and why]
Proposed Improvements
- [Improvement area]: [Concrete action]
- [Improvement area]: [Concrete action]
Which of these should I remember for future chats? [Ask user]
Skills Analysis
Skills used this session: [List skills]
For each skill:
skill-name:
- Self-check status: [✅ Has criteria / ❌ Missing]
- Token efficiency: [Observations]
- Proposed improvements:
- [Specific improvement]
- [Specific improvement]
Should I apply these improvements? [Ask user]
Pattern Detection
Repeated tasks identified:
- [Task pattern] (occurred X times)
Suggested skill: suggested-name
- Description: [What it would do]
Should I create this skill? [Ask user for confirmation]
Examples
User: reflect
Response:
## Session Review
**Accomplishments:**
- Updated comparison charts implementation plan
- Clarified data requirements for cycles.assessmentTool vs cycles.compare
- Committed plan to repository
**Mistakes/Issues:**
1. Initially didn't explain why cycles.compare was insufficient - required follow-up question
2. First plan draft missing reference cycle assessmentTool query
**Unclear Outputs:**
- Data structure explanation could have included visual diagram earlier
## Proposed Improvements
1. **Data explanations:** Include visual structure comparisons upfront when discussing API responses
2. **Plan completeness:** Always verify both sides of comparison features (reference + comparison)
3. **Proactive clarification:** Ask "should I explain why?" when making architectural decisions
Which of these should I remember for future chats?
## Skills Analysis
No skills were used in this session.
## Pattern Detection
No repeated task patterns detected yet.
Implementation Notes
- Review entire conversation history, not just recent messages
- Count actual occurrences of patterns (don't guess)
- Be specific about mistakes (include line numbers, quotes, context)
- Distinguish between mistakes and intentional iteration
- Focus on systemic issues, not one-off clarifications
- When suggesting new skills, check if similar skill already exists
- Token efficiency improvements should preserve all functionality
- Self-check criteria should be specific and measurable
Verification
Before presenting output:
- Have I reviewed the full chat history?
- Are all mistakes/friction points specific with examples?
- Are improvements concrete and actionable?
- Have I analyzed all skills used?
- Are my questions clear and answerable?
If any answer is no, iterate and improve (up to 5 times).