The Agent Skills Directory

[SAFE]: The skill implements a 'Library-First Protocol' that encourages the reuse of existing code from designated local paths (.claude/library/catalog.json and D:\Projects\*), which is a standard practice for development agents.
[SAFE]: Includes a 'Layout Restoration Pattern' that explicitly instructs the agent to research git history (using git log and git show) rather than using trial-and-error, which promotes reliable and traceable code changes.
[SAFE]: Defines clear 'User Frustration Signals' (e.g., 'stop', 'reverse all changes') that trigger an immediate rollback and strategy pivot, ensuring the agent remains responsive to user control.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process local project files and codebases. While this creates a surface for indirect prompt injection from malicious source code, the instructions include specific protocols for asset verification and session tagging that provide basic observability for these operations.

fix-bug