| name | deep-reflection |
| description | Self-improving skill that analyzes pipeline failures and user frustration, traces them to skill/agent gaps, and proposes concrete amendments. Fixes the immediate problem first, then reflects.
|
| imports | ["feedback-reinforcement","intellectual-honesty"] |
| invariants | ["Discussion checkpoint is MANDATORY before any skill/agent modification.","Analyzer agent is read-only. Only the orchestrator (this skill) writes files.","Max one reflection per conversation topic. Circuit breaker prevents loops."] |
Deep Reflection Skill
Self-improving skill that traces pipeline failures back to specific skills/agents and proposes
concrete amendments.
Invocation
/deep-reflection # Manual trigger — analyze the current issue
/deep-reflection --from-review # Called by /deep-review with findings as input
Step 1: Identify the Problem
- If
--from-review: Read the review findings
- Otherwise: Analyze the current conversation for friction points
- Categorize: pipeline failure, user frustration, quality gap, or efficiency issue
Step 2: Root Cause Analysis
Spawn 1 deep-reflection-analyzer agent (read-only):
- Traces the problem back to a specific skill or agent
- Identifies the gap: missing instruction, ambiguous wording, missing edge case
- Proposes a concrete amendment with before/after text
Step 3: Fix the Immediate Problem
Apply the amendment to the relevant skill/agent file.
Discussion checkpoint: Present the proposed amendment to the user before modifying anything.
Step 4: Feedback Entry
Save the correction to docs/feedback/ using the feedback-reinforcement skill's format.