The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references and directs the user to fetch installation instructions from the official inference-sh GitHub repository (e.g., https://raw.githubusercontent.com/inference-sh/skills/refs/heads/main/cli-install.md).
[COMMAND_EXECUTION]: The skill utilizes the belt command-line interface to execute web searches and data extraction tasks. The allowed-tools frontmatter correctly restricts the agent to only use the belt command within the Bash environment.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest arbitrary web content and feed it directly into LLM prompts (e.g., Claude), which is a common vector for indirect prompt injection.
Ingestion points: Untrusted content is fetched from external URLs via the tavily/extract and exa/extract apps in SKILL.md.
Boundary markers: The provided workflows use markdown-style tags like <search-results> and <content> to delimit the external data, but they lack explicit system-level instructions to ignore potential commands embedded within that data.
Capability inventory: The skill has the ability to execute network-enabled CLI tools (belt) and write data to local files (e.g., search_results.json).
Sanitization: There is no indication of content sanitization or filtering to remove malicious instructions from the extracted web data before it is processed by the agent.

web-search