The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill acts as a bridge for ingesting external PDF content into the agent's context, creating a potential surface for indirect prompt injection if a processed document contains malicious instructions.
Ingestion points: The skill instructions in SKILL.md detail how to read local files and pass their content to the crw_parse_file MCP tool or crw scrape CLI.
Boundary markers: There are no explicit instructions or delimiters defined to warn the agent to ignore instructions embedded within the extracted text.
Capability inventory: The skill is granted access to Bash and Read tools, which could be targets for manipulation via instructions found in a parsed document.
Sanitization: No sanitization or validation of the extracted document content is mentioned before the data is returned to the agent's context.

crw-parse