The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a variety of standard Unix shell commands including ls, find, grep, rm, mv, and mkdir to perform repository auditing and subsequent file management actions.
[COMMAND_EXECUTION]: Invokes external developer tools npx depcheck and python -m autoflake to identify unused dependencies and imports within the audited codebase.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted repository data (filenames, directory structures, and script contents) to generate an action plan for file deletion and relocation.
Ingestion points: Systematically scans the repository using ls and find in SKILL.md.
Boundary markers: None present; the agent treats the results of the filesystem scans as authoritative data for its logic.
Capability inventory: The skill includes the capability to delete files (rm), relocate files (mv), and execute shell scripts, providing a path for destructive actions if the agent is misled by malicious file patterns.
Sanitization: No explicit sanitization or validation of the findings is performed before presenting them as executable commands.
[PERSISTENCE]: Proposes the installation of a session hook in .claude/hooks/config.json that executes a bash command (find .claude/ -name ...) every time a new session starts.
[EXTERNAL_DOWNLOADS]: Fetches and executes the depcheck package from the npm registry via npx during the dependency audit phase.

repo-audit