The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the infsh CLI tool (allowed via Bash(infsh *)) to execute tasks on the vendor's platform.
[REMOTE_CODE_EXECUTION]: The primary function of this skill is to transmit and execute Python code on the remote inference.sh infrastructure. The environment is pre-configured with numerous libraries, including those for networking and browser automation.
[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection because it executes code provided in the code parameter, which an agent might generate based on untrusted input.
Ingestion points: The code field in the input JSON schema (SKILL.md).
Boundary markers: No specific delimiters or instructions are provided to the agent to treat the input code as potentially untrusted or to ignore embedded instructions.
Capability inventory: The remote environment provides a full Python 3.10 interpreter with capabilities for network access (e.g., requests, httpx), browser automation (e.g., selenium, playwright), and file system access within the sandbox, managed via the infsh tool.
Sanitization: The skill definition does not include logic for sanitizing or validating the Python code before execution.

python-executor