← ClaudeAtlas

vibe.capability-looplisted

Use when the same agent or tool failure repeats, or a missing capability blocks progress, to reproduce the cause and build a permanent repository capability.
su-record/vibe · ★ 0 · AI & Automation · score 78
Install: claude install-skill su-record/vibe
# Capability Loop — Failure → Capability Building > **Principle**: "Ask what capability is missing, not why it failed." Every failure is a missing tool, guardrail, abstraction, or piece of documentation. Build the capability so the failure NEVER recurs. ## When to Use | Trigger | Signal | |---------|--------| | Agent fails a task | Error, wrong output, timeout, confusion | | Same error occurs twice | Pattern of missing capability | | Agent asks clarifying question | Information not discoverable | | Review finds recurring issue | Systemic gap, not one-off mistake | | Manual intervention needed | Agent should have been self-sufficient | ## Core Flow ``` FAILURE → DIAGNOSE → CLASSIFY → BUILD → VERIFY → PERSIST ``` ### Step 1: DIAGNOSE — What Capability is Missing? Do NOT ask "why did it fail?" Instead ask: **"What would have prevented this failure?"** ``` Diagnosis questions (answer ALL): 1. TOOL: Did the agent lack a tool to accomplish the task? → Missing CLI command, missing API, missing utility function 2. GUARDRAIL: Did the agent do something it shouldn't have? → Missing lint rule, missing test, missing hook check 3. ABSTRACTION: Did the agent struggle with unnecessary complexity? → Missing helper, missing wrapper, missing shared module 4. DOCUMENTATION: Did the agent lack discoverable information? → Missing CLAUDE.md entry, missing code comment, missing type definition 5. FEEDBACK: Did the agent not know it was doing the wrong thing? → Missing er