| name | reflect |
| description | Reflect on experiment results, identify patterns, and propose improvements |
Reflect Skill
Purpose
Analyze experiment results, identify patterns/trends, diagnose failures, and propose concrete next steps.
When to Use
- After a batch of experiments completes
- When a run fails unexpectedly
- When results are surprising (good or bad)
- During periodic research reviews
Instructions
- Collect data: Gather metrics, logs, and configs from recent runs
- Compare: Look at how metrics changed across runs, identify best/worst configs
- Analyze patterns:
- What changed between successful and failed runs?
- Are there diminishing returns on certain hyperparameters?
- Are there surprising correlations?
- Diagnose issues if any runs failed:
- Was it a code bug, resource issue, or hyperparameter choice?
- Is the failure reproducible?
- Write reflection in this format:
## Reflection: [Topic]
### Summary of Runs
| Run | Config Change | Key Metric | Outcome |
|-----|--------------|------------|---------|
| ... | ... | ... | ... |
### Patterns Observed
- Pattern 1
- Pattern 2
### What Worked
- ...
### What Didn't Work
- ...
### Proposed Next Steps
1. [Concrete action with expected impact]
2. [Concrete action with expected impact]
- Create tasks for the proposed next steps on the Kanban board