The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to execute local Python scripts and system utilities. This includes running provided scripts like gap-analyzer.py as well as others mentioned in the workflow (e.g., detect-decomposition-targets.py, validate-references.py, audit-reference-depth.py). It also invokes developer tools such as ruff, pytest, git, and the GitHub CLI (gh).
[EXTERNAL_DOWNLOADS]: The skill automates repository workflows by using git push and gh pr create. It also employs gh pr merge --auto to automatically merge successfully validated reference changes, which facilitates an autonomous CI/CD style workflow for agent metadata.
[DATA_EXFILTRATION]: To perform its gap analysis, the skill reads agent and skill configurations from potentially sensitive local paths, specifically ~/.claude/agents/ and ~/.claude/skills/. While this is functional for its stated purpose, it involves reading the user's personal agent instructions and environment configurations.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core architecture. It ingests data from external agents and skills (which may be untrusted), processes this content to generate new reference files, and automatically commits these files to the repository. Malicious instructions embedded in a target agent's documentation could potentially influence the generation phase or the resulting automated PR operations.

reference-enrichment