The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes document content and comments, which are untrusted external inputs that could potentially contain malicious instructions.
Ingestion points: Document data is ingested using the read_docs tool, which fetches block content and comments as described in SKILL.md.
Boundary markers: The skill does not explicitly define delimiters to separate untrusted content from system instructions in the agent's context.
Capability inventory: The agent can modify document structure, delete content, and add comments through tools like update_doc and add_comment.
Sanitization: The instructions focus on structural validation (e.g., block ID format and position checks) rather than semantic filtering of document text content.

docs-collaborator