The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/content_extract.py executes a local Python script mineru_parse_documents.py (part of the mineru-extract skill) using subprocess.run. The user-supplied URL is passed as a command-line argument. The implementation uses a list-based argument format, which safely avoids shell injection risks.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because its primary function is to ingest and process untrusted data from external websites. Ingestion points: The --url argument in scripts/content_extract.py and the web_fetch tool mentioned in SKILL.md. Boundary markers: No explicit boundary markers or 'ignore' instructions are added to the extracted markdown before it is returned to the agent. Capability inventory: The skill has the capability to execute local sub-processes (subprocess.run) and return the output of external web fetches to the agent context. Sanitization: No sanitization or filtering of the extracted markdown content is performed before returning it to the agent.

content-extract