The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest large amounts of external data from user-provided files and shared workspace integrations.
Ingestion points: Directly reads shared documents, files, and content from team channels (e.g., Slack, Google Drive, SharePoint) as specified in the SKILL.md instructions.
Boundary markers: The instructions do not define explicit delimiters to separate external content from internal logic, relying on the agent's contextual understanding.
Capability inventory: The agent uses create_file and str_replace to manage document content and can spawn sub-agents to test the final draft.
Sanitization: There are no explicit instructions for filtering or sanitizing instructions that might be embedded in the ingested documents.
[COMMAND_EXECUTION]: The skill uses specific file manipulation tools to draft and edit documents.
Evidence: Instructions explicitly direct the agent to use create_file to initialize document artifacts and str_replace to perform granular edits on the sections during the optimization phase.
[EXTERNAL_DOWNLOADS]: The skill interacts with external document and communication platforms via official connectors.
Evidence: References to reading data from Google Drive, SharePoint, Slack, and Microsoft Teams. These downloads are facilitated through standard platform integrations and MCP servers intended for legitimate context gathering.

doc-coauthoring