The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill interacts with external data from third-party providers (e.g., GitHub, Slack) which could contain malicious content. The skill instructions proactively mitigate this risk by requiring the agent to treat external content as data, not instructions.
Ingestion points: Data retrieved from external providers through the execute_composio_tool function.
Boundary markers: Explicit instructions in references/security.md state that provider descriptions and records are untrusted data and must not influence action selection or authorization.
Capability inventory: The skill has the ability to execute writes to external services via execute_composio_tool and perform destructive actions like toolkit disconnection.
Sanitization: Instructions mandate the redaction of sensitive credentials and the truncation of large payloads to prevent data exposure.
[EXTERNAL_DOWNLOADS]: The skill references an external Model Context Protocol (MCP) server for its core functionality.
Evidence: The configuration in agents/openai.yaml specifies a connection to https://console.mermail.app/mcp using streamable_http transport.

mermail-composio