The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted web content via search and crawl tools, which is the primary intended purpose of the skill. This creates a surface for potential indirect prompt injection where instructions embedded in external content could attempt to influence agent behavior.
Ingestion points: Results from search(), extract(), crawl(), and research() methods (defined in SKILL.md and reference files) return raw or chunked content from external URLs.
Boundary markers: The documentation demonstrates how to access and print results but does not enforce specific delimiter usage or prompt wrapping techniques within the examples.
Capability inventory: The agent is provided with tools for web search, URL discovery, and autonomous research synthesis using the Tavily API.
Sanitization: The reference documentation focuses on data retrieval; sanitization of external content is expected to be managed by the implementation layer or the AI model's own safety guardrails.
[SAFE]: The skill adheres to security best practices for credential management by recommending the use of environment variables and the getpass utility for API keys. All listed dependencies are well-known, official packages from established registries (NPM and PyPI). All referenced URLs point to official vendor domains or trusted integration providers.

tavily-sdk