The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest significant amounts of external context from user 'info dumps', shared document links, and team messaging channels (via MCP integrations like Slack or Teams). This creates an attack surface where malicious instructions hidden in these external sources could theoretically influence the agent's drafting behavior. However, this risk is inherent to the primary purpose of a documentation assistant.
Ingestion points: User-provided text, uploaded files, and external content retrieved via platform integrations (Google Drive, SharePoint, etc.) in SKILL.md.
Boundary markers: The skill does not explicitly define delimiters for separating user-provided context from internal instructions.
Capability inventory: The skill uses create_file and str_replace to manage the document artifact and invokes sub-agents for testing.
Sanitization: No specific sanitization or filtering logic is provided for the ingested content.
[DYNAMIC_EXECUTION]: In the 'Reader Testing' stage, the skill uses a multi-agent approach where it invokes a 'fresh Claude' (sub-agent) to verify the document's clarity and answer predicted reader questions. This is a functional use of the platform's sub-agent capabilities to ensure document quality without context bleed.

doc-coauthoring