The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes helper scripts (setup_composio_mcp.sh) that modify local filesystem configuration files for various AI agent platforms (e.g., Claude, Cursor, Devin) and execute third-party CLI tools like openclaw to manage integrations.\n- [DYNAMIC_EXECUTION]: The skill facilitates the use of composio run, which allows for the execution of arbitrary JavaScript or TypeScript snippets within the agent's environment. Furthermore, the included setup script utilizes an embedded Python script to programmatically alter JSON configurations on the host.\n- [INDIRECT_PROMPT_INJECTION]: By integrating with over 1000 external applications such as Gmail, Slack, and GitHub, the skill creates a surface for indirect prompt injection attacks. Malicious instructions could be embedded in the content retrieved from these services.\n
Ingestion points: Tools like GMAIL_FETCH_EMAILS and GITHUB_LIST_REPOSITORY_ISSUES ingest untrusted content into the agent's context.\n
Boundary markers: No specific boundary markers or instructions to ignore embedded commands were identified in the provided materials.\n
Capability inventory: The skill possesses extensive capabilities including file system modification, tool execution, and network access to vendor APIs.\n
Sanitization: There is no evidence of explicit sanitization or validation of the data fetched from external toolkits before it is processed by the agent.

composio-mcp