The Agent Skills Directory

[COMMAND_EXECUTION]: The script executes shell commands to perform its audit duties.\n
Evidence: Uses git -C "$marketplace_dir" fetch origin and git reset --hard origin/main to update local copies of marketplace repositories.\n
Evidence: Executes ps aux via Python subprocess to inspect the command-line arguments of running Node.js processes for version verification.\n- [DATA_EXFILTRATION]: The skill accesses system metadata and local configuration files to perform its audit.\n
Evidence: Reads ~/.claude/plugins/installed_plugins.json to identify installed plugins and their expected versions.\n
Evidence: Scans the process list (ps aux) which exposes information about all running processes on the system to the script.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through external marketplace metadata.\n
Ingestion points: Processes marketplace.json from remote git repositories managed by the marketplace vendor.\n
Boundary markers: The script does not utilize explicit boundary markers or instructions to Claude to ignore potential instructions embedded in the plugin names or version strings.\n
Capability inventory: The skill identifies and suggests high-impact remediation actions like rm -rf (deleting directories) and kill (terminating processes).\n
Sanitization: The script does not sanitize or validate strings read from the remote marketplace.json before outputting them for AI interpretation.

refresh-plugin-cache