What problem does it solve? When users correct an AI coding agent's behavior, the fix often lands in the nearest instruction file instead of the source that actually owns the behavior, so the same mistake recurs. This Skill turns feedback into durable improvements by choosing the right enforcement level and scope before editing anything. ## Core Features & Use Cases - Enforcement ladder: Prioritizes hooks and structural gates first, then skills, then project code, with instruction files as the last resort. - Scope ladder: Determines whether a rule applies cross-agent, cross-project, per-repo, or per-subproject before choosing a target. - Marketplace routing: Detects plugin marketplaces via manifest files and routes fixes to hook scripts, SKILL.md sources, or plugin metadata instead of runtime caches. - Pruning discipline: Deletes, merges, and simplifies duplicated or stale rules instead of only appending new ones. - Use Case: A user corrects the agent's commit message style for the third time. Instead of adding another line to CLAUDE.md, the Skill routes the fix into the git-discipline hook source so the rule is enforced structurally across all repos. ## Quick Start Tell the agent to apply this feedback as a durable fix: the agent keeps forgetting to run tests before committing, route it to the right source.
