The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends installing the tavily-cli using uv or pip. These are standard installation methods for a well-known service's official tool.
[COMMAND_EXECUTION]: The skill instructions involve executing the tvly CLI command with various arguments to perform web searches and check status. It includes a specific warning against including real API keys in command arguments, which is a positive security practice.
[PROMPT_INJECTION]: The skill includes a 'Verification Protocol' and 'Anti-Patterns' section that explicitly warns the agent to treat external content as untrusted data and verify claims. While these are good defensive measures, the skill inherently processes untrusted data from the web, creating a surface for indirect prompt injection.
[PROMPT_INJECTION]: The 'Indirect Prompt Injection' surface is identified:
Ingestion points: Search results, snippets, and raw page content fetched via tvly search (SKILL.md).
Boundary markers: The 'Verification Protocol' requires treating external content as untrusted data and quoting material (SKILL.md).
Capability inventory: The skill can execute shell commands (tvly), which could be abused if the agent is tricked by malicious search results (SKILL.md).
Sanitization: No explicit programmatic sanitization is defined; the skill relies on the agent's internal guardrails and the 'Verification Protocol' instructions (SKILL.md).

tavily-search