The Agent Skills Directory

[PROMPT_INJECTION]: The skill demonstrates workflows that are susceptible to indirect prompt injection. It extracts content from arbitrary URLs and interpolates that untrusted data directly into prompts for large language models.
Ingestion points: Data retrieved via tavily/extract and exa/extract in SKILL.md is fed into subsequent LLM calls.
Boundary markers: The provided examples lack delimiters or specific instructions to the agent to treat the extracted content as untrusted data.
Capability inventory: The skill utilizes the infsh tool to perform both data retrieval and model invocation, creating a combined read-and-process chain.
Sanitization: There is no mention of sanitizing or validating the content extracted from external websites before it is processed by the AI.
[EXTERNAL_DOWNLOADS]: The documentation provides a link to an external shell script for installing the required infsh CLI tool.
[COMMAND_EXECUTION]: The skill is designed to execute the infsh command-line utility to interact with various search and AI applications. This tool is explicitly listed in the allowed-tools configuration.

web-search