miles-rl-training
Pass
Audited by Gen Agent Trust Hub on Apr 30, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill describes workflows for processing external data files (e.g.,
--prompt-data /path/to/data.jsonl) during training and rollout. This creates an indirect prompt injection surface where malicious content within the training data could attempt to influence the agent's behavior or the model's training outcomes. - Ingestion points: Data entering via the
--prompt-dataargument inSKILL.mdworkflows. - Boundary markers: None explicitly specified for the training data inputs.
- Capability inventory: Execution of shell commands and Python training scripts that interact with local files and hardware (SKILL.md, references/api-reference.md).
- Sanitization: No specific sanitization or validation of the input data content is described.
- [EXTERNAL_DOWNLOADS]: The skill references external resources for installation, including a GitHub repository (
github.com/radixark/miles) and a Docker image (radixark/miles:latest). These resources appear consistent with the identified author and the framework's stated purpose. - [COMMAND_EXECUTION]: The skill provides numerous shell commands for environment configuration (e.g.,
exportcommands), dependency management (pip install), and running complex training scripts (python train.py).
Audit Metadata