Turn repeated agent mistakes into lasting fixes.
What it does
Identifies recurring agent failures and token-spend drivers, then proposes targeted improvements for user approval.
When to use it
Use for recurring correction analysis, rejection-rate trends, token-spend audits, or the scheduled maintenance nudge.
How to use it
Give it the system's work folder; it reports correction patterns and metrics, asks for per-item decisions, and applies only approved changes.
What you provide
- An existing project
Uses
Good to know
- starts helper agents, number unstated
- cannot be undone: ~/.claude/agents/; session memory directory
- works out of sight
What you need · 2
Requires Python to run the maintenance-cadence marking command.
Routes patterns supported only by overlays to the curation workflow instead of drafting their edits here; installation requirements are unstated.
Access · 5
This skill
~/.claude/agents/
Edits your canonical agent files (irreversible)
session memory directory
Edits approved feedback memories (irreversible)