simpo-training
Pass
Audited by Gen Agent Trust Hub on Sep 16, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill references and downloads resources from trusted organizations.\n
- Fetches the
alignment-handbookrepository from Hugging Face's official GitHub account inSKILL.md.\n - References datasets hosted on Hugging Face Hub from reputable organizations like Meta, Anthropic, and NVIDIA in
references/datasets.md.\n- [COMMAND_EXECUTION]: The skill provides instructions for running standard shell commands for environment setup and training.\n - Includes commands for package installation via
pipand environment management viacondainSKILL.md.\n - Includes training launch commands using the
accelerateframework inSKILL.md.\n- [INDIRECT_PROMPT_INJECTION]: The skill describes the ingestion of external preference datasets for model alignment.\n - Ingestion points: The
dataset_mixerconfiguration inSKILL.mdallows loading external datasets from Hugging Face.\n - Boundary markers: The skill does not explicitly mention prompt boundary markers but instructions are for structured training data.\n
- Capability inventory: The skill utilizes
accelerate launchfor subprocess execution and writes model outputs to the local file system.\n - Sanitization: The
references/datasets.mdfile provides examples of data quality filtering and validation scripts to ensure data integrity before training.
Audit Metadata