The Agent Skills Directory

[SAFE]: The skill provides a legitimate interface for web research and content aggregation using tools from a well-known service provider.
[COMMAND_EXECUTION]: Uses the Bash tool to execute commands via the belt CLI. The scope is restricted through the allowed-tools frontmatter to ensure the agent only uses the intended toolset.
[EXTERNAL_DOWNLOADS]: References installation scripts and CLI packages from the official inference-sh GitHub repository and NPM, which are recognized sources for this functionality.
[PROMPT_INJECTION]: The skill processes untrusted web content, creating a surface for indirect prompt injection.
Ingestion points: belt app run tavily/extract and exa/extract in SKILL.md.
Boundary markers: Demonstrated use of XML tags (e.g., <search-results>) to delimit external content.
Capability inventory: Shell access is restricted to the belt CLI via allowed-tools.
Sanitization: Not explicitly implemented in the skill instructions.

web-search