humblelisted
Install: claude install-skill x-mesh/xm
<Purpose>
x-humble is a structured retrospective system where the user and agent learn from failures together.
The **retrospective process itself** is the core, not rule generation. Rules are merely a byproduct.
"humble = knowing you can be wrong"
</Purpose>
<Use_When>
- At session end: "what did I get wrong?", "let's reflect", "retrospect", "reflect"
- When the same mistake repeats: "this again", "why does this keep happening"
- After project completion: "what did we learn?", "how should we do it next time?"
- When the user has said "that's not right" multiple times
</Use_When>
<Do_Not_Use_When>
- When a problem needs to be solved right now (use x-solver)
- When strategic analysis is needed (use x-op)
- For simple feedback storage (use Claude Code built-in memory)
</Do_Not_Use_When>
# x-humble — Structured Retrospective
A structured retrospective to learn from failures together. The user learns, the agent learns.
## Model Routing
| Phase | Model | Reason |
|-------|-------|--------|
| Phase 0 (CHECK-IN: read prior commitments) | **haiku** (Agent tool) | File read + status display |
| Phase 1 (RECALL: aggregate git/work history) | **haiku** (Agent tool) | Mechanical aggregation |
| Phase 2-5 (IDENTIFY/ANALYZE/ALTERNATIVE/COMMIT) | **sonnet** | Reasoning, bias analysis, judgment |
**Guardrail**: never haiku for ANALYZE or ALTERNATIVE — retrospective quality requires reasoning. Downgrading these silently degrades learning capture.
## Mode Detection
Read mode from `.xm