What problem does it solve? Toolguard and Claude Code permission configurations grow over time into redundant, mis-levelled, and confusing rule sets that are hard to audit and risky to clean up by hand. This Skill turns that cleanup into a guided, evidence-backed review where nothing is ever applied without explicit per-item consent. ## Core Features & Use Cases - Analyzer-driven evidence gathering: Runs the deterministic toolguard-maintain and toolguard-audit analyzers to find redundancies, consolidation candidates, broadenings, and cross-layer interactions, then reasons over the JSON findings in structured passes. - Family-grouped proposals with certification: Groups every rule into command families, targets the right config level for each change, blocks cross-level welds, renders paste-ready TOML, and certifies it via staged audits and corpus replay before any write. - Consent-gated application with decision memory: Applies only user-approved changes via --apply --write or guided self-edit, and records decisions in-file (# toolguard: annotations) and in a sidecar ledger so periodic runs stay quiet. - Use Case: Before pushing, ask for a periodic maintenance checkpoint on your project; the Skill surfaces only new or changed permission rules, proposes safe consolidations, and applies just the ones you approve. ## Quick Start Ask the assistant to clean up and consolidate the toolguard permission rules for the current project, reviewing each proposed change before anything is applied.
