The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructs the agent to check the contents of ~/.pytrio/config.toml to verify the presence of an api_key. Accessing and inspecting local configuration files for stored credentials constitutes a security risk.
[COMMAND_EXECUTION]: The skill provides instructions to execute shell commands, specifically 'trio login -k <API_KEY>', where the API key is a parameter provided by the user. This pattern involves command execution with externally-sourced input.
[PROMPT_INJECTION]: The skill establishes an indirect prompt injection surface by providing code examples (e.g., in examples/01_train_sft.py) that process training data consisting of prompts and completions from potentially untrusted sources. This data is ingested without sanitization or boundary markers and subsequently used in remote API calls for training and inference.
[DATA_EXFILTRATION]: The skill utilizes the PyTRIO SDK to transmit data, including training samples, model weights, and performance metrics, to the remote domain pytrio.cn.
[EXTERNAL_DOWNLOADS]: The documentation directs the installation of several packages, including pytrio, transformers, torch, numpy, datasets, and modelscope, from public package registries.

pytrio