The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the execution of arbitrary Python code in a remote sandboxed environment via the Together AI Code Interpreter API. This includes functionality for stateful sessions and chart generation.
[EXTERNAL_DOWNLOADS]: The skill involves installing the together Python SDK and using npx for TypeScript execution, which can trigger package downloads from official registries.
[DATA_EXFILTRATION]: The skill's ability to upload files to a remote environment creates a potential path for data exfiltration if an agent is instructed to send sensitive local data to the API.
[PROMPT_INJECTION]: An indirect prompt injection surface exists as the skill processes stdout and stderr from the remote environment without sanitization.
Ingestion points: Remote execution outputs are ingested in execute_with_session.py and execute_with_session.ts.
Boundary markers: None provided in the implementation scripts.
Capability inventory: The skill allows local file writing and network communication with the Together AI API.
Sanitization: No filtering or instruction-ignoring delimiters are applied to the outputs received from the remote sandbox.

together-code-interpreter