| name | Reflection |
| description | Structured self-evaluation before delivering work. Catches blind spots and improves quality through multi-lens critique. |
| version | 1.0.0 |
When to Use
Before delivering important work, pause for structured self-evaluation. Catches blind spots, improves quality, builds trust.
Triggers:
- Complex deliverables (architecture, strategy, analysis)
- High-stakes outputs (production code, client-facing docs)
- Uncertainty about quality ("is this actually good?")
- After extended work sessions (tunnel vision risk)
The Reflection Process
1. Distance — Step back mentally. Pretend you're reviewing someone else's work.
2. Multi-lens evaluation:
- Correctness: Does it actually solve the problem stated?
- Completeness: What's missing? Edge cases? Assumptions?
- Clarity: Would someone else understand this immediately?
- Robustness: What could go wrong? What breaks this?
3. Steel-man critique — Find the strongest objection to your work. Address it.
4. Honest assessment — Rate confidence 1-10. Below 7? State what would raise it.
Reflection Depth
| Situation | Depth |
|---|
| Quick answer | 10 seconds, one lens |
| Standard task | 30 seconds, all lenses |
| Critical delivery | 2 minutes, full critique |
Red Flags to Catch
- First solution accepted without alternatives considered
- Assumptions not validated with user
- Edge cases hand-waved
- Complexity added without clear benefit
- "It works" without understanding why
Output Format
After reflection, either:
- Deliver with confidence — reflection confirmed quality
- Improve then deliver — found issues, fixed them
- Flag uncertainty — "I'm 6/10 confident because X, want me to dig deeper?"
See dimensions.md for domain-specific evaluation criteria.
See prompts.md for self-reflection questions by task type.
Related: diverge (multiple perspectives), brainstorm (idea generation)
