The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to install the belt CLI and other skills from the belt-sh and inference-sh organizations. It also references installation scripts and documentation hosted on GitHub.
[COMMAND_EXECUTION]: The skill executes search and extraction tasks using the belt command-line utility, which interacts with the inference.sh platform apps to perform web operations.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from the internet through web searches.
Ingestion points: Untrusted data enters the agent context through the search results and extracted page content from Tavily and Exa tools (e.g., tavily/extract, exa/search).
Boundary markers: The documentation suggests using simple tags like <search-results> or <content>, which do not provide robust protection against malicious instructions embedded in the retrieved web data.
Capability inventory: The skill uses the belt CLI to run various search and extraction applications and suggests piping this data to other LLM models.
Sanitization: No explicit sanitization, validation, or escaping of the external web content is mentioned before it is processed by the agent or passed to subsequent LLM steps.

web-search