The Agent Skills Directory

[COMMAND_EXECUTION]: The skill directs the agent to generate and run local Python scripts to handle complex multi-step workflows, such as browser session management with Playwright and chunked media uploads for Twitter. This creates a surface for dynamic code execution based on the provided templates.\n- [COMMAND_EXECUTION]: The skill provides numerous examples for using curl to interact with the internal gateway API (http://composio-gateway.flycast) and external endpoints.\n- [EXTERNAL_DOWNLOADS]: Operational flows depend on several external Python libraries, specifically composio_client, playwright, httpx, and requests.\n- [EXTERNAL_DOWNLOADS]: The skill performs network requests to external service domains including composio-gateway.fly.dev and connect.browserbase.com.\n- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its core function of ingesting untrusted data from external APIs like Gmail and GitHub.\n
Ingestion points: Untrusted data enters the agent context through tools like GMAIL_FETCH_EMAILS, GITHUB_GET_REPOSITORY_CONTENT, and GOOGLEDOCS_GET_DOCUMENT_PLAINTEXT as described in SKILL.md.\n
Boundary markers: The instructions lack explicit delimitation or 'ignore instructions' warnings when processing retrieved external content.\n
Capability inventory: The skill allows the agent to execute shell commands via curl and run generated Python scripts, which could be exploited by malicious data.\n
Sanitization: There is no documentation or instruction for sanitizing or validating external content before the agent processes it.

composio