The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documentation includes instructions to install several external packages from public registries (PyPI and NPM), including composio, claude-agent-sdk, openai-agents, and @composio/core. These are official libraries for the Composio service and are standard for this type of integration.
[PROMPT_INJECTION]: The skill is designed to ingest data from untrusted external sources such as emails, chat messages, and issue trackers, creating a potential surface for indirect prompt injection.
Ingestion points: Untrusted data enters the agent context from connected services (Gmail, Slack, GitHub, etc.) as noted in SKILL.md.
Boundary markers: The provided code examples and instructions do not specify the use of delimiters or boundary markers for ingested content.
Capability inventory: The skill provides extensive capabilities to take actions (send emails, create issues, update databases) across a wide range of categories (Email, Chat, Dev, CRM, etc.).
Sanitization: There is no mention of sanitization or validation of content fetched from external tools before it is processed by the agent.
[DATA_EXFILTRATION]: While the skill is intended to transmit data to external services, it correctly instructs users to manage sensitive credentials using environment variables (COMPOSIO_API_KEY) and relies on OAuth for service-specific authorizations, which limits unauthorized data exposure.

connect