The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the infsh command-line tool via Bash to interface with the remote execution environment. This is the primary method for running applications on the vendor's platform.
[REMOTE_CODE_EXECUTION]: The skill is designed to execute arbitrary Python code provided in the code parameter. This code is processed in a remote sandboxed environment provided by inference.sh.
[DATA_EXFILTRATION]: The execution environment includes pre-installed networking libraries such as requests, httpx, aiohttp, and browser automation tools like selenium and playwright. These tools enable the executed scripts to perform network operations and communicate with external web services.
[PROMPT_INJECTION]: The skill provides a surface for indirect prompt injection as it executes Python code provided through the agent's input.
Ingestion points: The code field defined in the input schema within SKILL.md is the primary entry point for untrusted instructions.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to sanitize or ignore embedded instructions within the input data.
Capability inventory: The environment allows full Python execution with access to network libraries and the ability to write files to the outputs/ directory.
Sanitization: No validation or escaping of the provided Python code is performed before execution.

python-executor