The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external files (such as PDFs, DOCX, or HTML files) which presents an indirect prompt injection surface if those documents contain adversarial text intended to hijack the agent's behavior.
Ingestion points: File paths supplied to the firecrawl parse command in SKILL.md.
Boundary markers: Absent; there are no delimiters or instructional boundaries defined to prevent the agent from obeying instructions embedded within the parsed text.
Capability inventory: Shell execution capabilities via Bash(firecrawl *) and Bash(npx firecrawl-cli *) are used to run the tool and write data to disk.
Sanitization: Absent; the content is parsed or summarized directly without filtering or verification mechanisms.
[COMMAND_EXECUTION]: The skill invokes system commands using the Bash tool to run the firecrawl binary or execute firecrawl-cli via npx.

firecrawl-parse