The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill facilitates the execution of user-supplied Python code on Together AI's remote sandbox environment. This is the core functionality of the skill and is implemented through official API endpoints (https://api.together.ai/tci/execute) using the vendor's SDK.
[EXTERNAL_DOWNLOADS]: The skill utilizes official vendor libraries, including the together Python package (>=2.0.0) and together-ai Node.js package. It also allows for runtime installation of additional packages within the sandbox via "!pip install", which is an expected feature for this service.
[COMMAND_EXECUTION]: The skill supports executing shell commands within the remote sandbox environment using the "!" prefix (e.g., "!python myscript.py"), which is standard for Jupyter-like execution environments.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface because it executes code generated or passed through an agent.
Ingestion points: The "code" parameter in the execute_code function and API requests within scripts/execute_with_session.py and scripts/execute_with_session.ts.
Boundary markers: None identified in the provided client-side scripts; the skill relies on the calling agent to properly delimit instructions from data.
Capability inventory: Full Python 3 execution, network access (within sandbox limits), and shell command execution via the "!" operator.
Sanitization: The skill is designed to execute the provided string directly without sanitization, as is standard for remote code interpreter tools.

together-sandboxes