The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill enables an indirect prompt injection surface by fetching content from arbitrary URLs and passing it into LLM prompts.
Ingestion points: Data is ingested from external web pages via the tavily/extract and exa/extract tools.
Boundary markers: Example workflows in the documentation use <search-results> and <content> tags as delimiters for external data.
Capability inventory: The skill is granted Bash access, specifically restricted to the belt command prefix.
Sanitization: The instructions do not define any sanitization or validation steps for content extracted from the web.
[EXTERNAL_DOWNLOADS]: The skill references the installation of the belt CLI tool via npx and points to resource files in the inference-sh GitHub repository. These are legitimate resources for the skill's stated purpose.
[COMMAND_EXECUTION]: The skill uses the Bash tool to execute belt CLI commands. Access is properly scoped to the belt application via the allowed-tools frontmatter field.

web-search