The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The primary function of this skill is to execute arbitrary Python code on a remote server (inference.sh) using the belt CLI tool. This is the intended behavior for providing sandboxed code execution capabilities to an agent.
[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it ingests and executes code provided in the code input parameter. An agent that generates this code based on untrusted external data (such as web content) could be influenced to perform unintended actions within the sandbox.
Ingestion points: The code field in the input JSON schema (SKILL.md).
Boundary markers: None explicitly defined in the prompt template, though the skill notes the environment is sandboxed.
Capability inventory: The skill can perform network operations (requests, scrapy), file manipulation (pillow, moviepy), and arbitrary computation via the belt CLI tool.
Sanitization: Not explicitly documented; the skill relies on the remote infrastructure's sandboxing for safety.
[EXTERNAL_DOWNLOADS]: The skill references and encourages the installation of the belt CLI tool from a public registry and fetches configuration/installation documentation from the provider's official GitHub repository (inference-sh organization).

python-executor