The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external, potentially untrusted sources such as URLs, shared documents, and team communication channels. This creates a surface for indirect prompt injection where malicious instructions embedded in those external sources could influence the agent's behavior during the co-authoring process.
Ingestion points: The skill fetches content from user-provided links, shared document platforms, and team chat threads mentioned during the 'Context Gathering' stage in SKILL.md.
Boundary markers: There are no specific instructions or delimiters provided to the agent to distinguish between the external data and the system instructions, increasing the risk that the agent might follow instructions found within the fetched content.
Capability inventory: The agent is authorized to use tools for fetching web content, reading document files, and creating new .docx files via the word-documents skill.
Sanitization: The instructions do not specify any validation or sanitization steps for the content retrieved from external sources.
[EXTERNAL_DOWNLOADS]: The workflow explicitly instructs the agent to fetch content from external URLs and shared document links provided by the user to gather context. While this is part of the intended functionality, it involves downloading data from remote locations into the agent's active context.

doc-coauthoring