The Agent Skills Directory

REMOTE_CODE_EXECUTION (CRITICAL): The instruction to use curl -fsSL https://cli.inference.sh | sh is a confirmed remote code execution pattern. This pattern allows an untrusted third-party server to execute arbitrary commands on the host machine without user verification or oversight.
COMMAND_EXECUTION (HIGH): The skill is entirely dependent on executing commands via the Bash tool (infsh *). This provides an expansive capability set that could be abused if the inputs to these commands are not strictly controlled.
EXTERNAL_DOWNLOADS (HIGH): The skill relies on external binaries and scripts downloaded from inference.sh, which is not a recognized trusted source in the defined security policy.
Indirect Prompt Injection (LOW): The skill ingests data from arbitrary URLs and processes it through an LLM, creating a surface for injection attacks.
Ingestion points: Data entering through tavily/extract and exa/extract commands in SKILL.md.
Boundary markers: Absent. The documentation shows direct interpolation of search/extract results into LLM prompts without delimiters or warnings.
Capability inventory: Bash capability used for file writing (> search_results.json) and network communication via the infsh CLI.
Sanitization: None detected in the provided workflow examples.

web-search