torchforge-rl-training

Pass

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external datasets (e.g., 'openai/gsm8k' mentioned in configuration examples) which may contain adversarial content designed to influence agent behavior.\n
  • Ingestion points: Dataset paths in configuration files (SKILL.md).\n
  • Boundary markers: Absent; there are no instructions or delimiters provided to isolate untrusted dataset content from agent instructions.\n
  • Capability inventory: The skill uses 'python -m' to execute training and inference processes that process this data.\n
  • Sanitization: No sanitization or validation logic is described for the incoming data.\n- [EXTERNAL_DOWNLOADS]: The skill references local installation scripts ('./scripts/install.sh' and './scripts/install_rocm.sh') for environment setup that are not included in the provided file set.\n
  • This prevents security verification of the commands being executed or any third-party resources they may download during the setup process.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 17, 2026, 07:53 PM
Security Audit — agent-trust-hub — torchforge-rl-training