The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core functionality of reading and processing external data.
Ingestion points: The workflow retrieves content from external sources including shared document links (Google Drive, SharePoint), messaging channel threads (Slack, Teams), and user-provided template files as described in Stage 1 and Stage 2.
Boundary markers: The instructions lack explicit delimiters or specific 'ignore embedded instructions' warnings for the data ingested from external channels, which could allow malicious instructions in those sources to influence agent behavior.
Capability inventory: The skill utilizes file-system tools such as create_file and str_replace to generate and modify document content based on the gathered context. It also leverages network-based integrations (connectors/MCP servers) to access external data.
Sanitization: There are no instructions for the agent to sanitize, escape, or validate the integrity of the content retrieved from external messaging or document platforms before drafting.

doc-coauthoring