The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions to use local utilities like ls and grep to find and search through existing skill files stored in the user's local environment (~/.claude/skills/).
[PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks as it is designed to ingest and formalize content from potentially untrusted conversation sessions. * Ingestion points: The skill analyzes the current conversation history and session context to identify learning candidates. * Boundary markers: There are no technical delimiters used to separate the session data from the generated skill instructions. * Capability inventory: The skill possesses Write access to the filesystem (to save skills) and WebSearch capabilities. * Sanitization: The skill relies on a manual quality assessment and requires explicit user confirmation (Step 2) before any data is saved to the persistent skill storage.

skill-extractor