The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches machine learning dependencies (trl, transformers, datasets, peft, accelerate, torch) and training datasets (trl-lib, openai, argilla) from official registries and the Hugging Face Hub. These are well-known and trusted sources for AI research and development.
[COMMAND_EXECUTION]: Provides templates and instructions for executing model fine-tuning tasks via the TRL CLI and Python scripts. These commands are intended for local model training and follow the library's official guidelines.
[INDIRECT_PROMPT_INJECTION]: The skill processes external datasets for reinforcement learning. While this constitutes an attack surface for indirect prompt injection, it is the primary intended functionality of the skill and uses standard data loading utilities. Ingestion points: 'load_dataset' calls in SKILL.md and training templates. Boundary markers: Not explicitly defined in training snippets. Capability inventory: Includes file writing for model checkpoints and network communication for logging. Sanitization: Relies on standard tokenization and library-level processing.

trl-fine-tuning