The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches content from external Google Docs and processes it locally, which presents a risk of indirect prompt injection. \n
Ingestion points: The skill uses mcp__google-docs__get_doc_content to retrieve the body of Google Documents (SKILL.md). \n
Boundary markers: No boundary markers or delimiters are specified when writing the document content to local files, making it difficult for the agent to distinguish between instructions and data. \n
Capability inventory: The skill utilizes Write and Bash tools, which could be exploited if malicious instructions within a document are obeyed. \n
Sanitization: While filename sanitization is performed (reference.md), there is no evidence of sanitization or filtering for the actual document content before it is saved or processed.

import-gdoc