The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a surface for Indirect Prompt Injection (Category 8) due to its design for documentation co-authoring.
Ingestion points: The workflow explicitly prompts the agent to read context from external sources including Slack, Microsoft Teams, Google Drive, SharePoint, arbitrary URLs, and uploaded files during the 'Context Gathering' stage.
Boundary markers: There are no explicit delimiters or system instructions provided to the agent to distinguish between the user's workflow commands and potentially malicious instructions embedded within the ingested documents or messages.
Capability inventory: The skill utilizes create_file and str_replace tools to generate and modify files or artifacts based on the gathered context. It also has the ability to spawn sub-agents for 'Reader Testing'.
Sanitization: The instructions do not specify any validation or sanitization steps for data pulled from third-party integrations before it is processed or used to generate file content.
[COMMAND_EXECUTION]: The skill instructs the agent to use file system tools (create_file, str_replace) to manage the document drafting process. This behavior is documented, purposeful, and restricted to the documentation task defined in the workflow.

doc-coauthoring