The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted local files (e.g., PDF, DOCX, HTML) and converts them to markdown for the agent to inspect. This creates a vulnerability surface for indirect prompt injection, where instructions hidden inside a document could attempt to hijack the agent's logic.
Ingestion points: Untrusted local files processed via firecrawl parse in SKILL.md.
Boundary markers: Absent. The instructions do not define delimiters or provide warnings for the agent to ignore instructions embedded within the parsed document content.
Capability inventory: The skill is authorized to use the Bash tool, which can execute commands and perform file operations.
Sanitization: Absent. There is no evidence of content filtering or sanitization during the conversion process to prevent the agent from obeying malicious instructions in the output markdown.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute firecrawl and npx firecrawl-cli commands. The use of the wildcard * in allowed-tools permits the agent to execute any subcommand or argument provided by the Firecrawl CLI, which is consistent with its intended use as a vendor-provided tool.

firecrawl-parse