debug-graph-breaks
Pass
Audited by Gen Agent Trust Hub on Jun 29, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses shell commands to run user-provided Python scripts with environment variables like
TORCH_LOGSto identify graph breaks.\n - Evidence: Instructions in
SKILL.mdto runTORCH_LOGS=\"graph_breaks\" python your_script.py.\n- [EXTERNAL_DOWNLOADS]: The skill retrieves documentation from the vendor's official websites to provide context-specific guidance for fixing compilation errors.\n - Evidence: Fetches from
https://meta-pytorch.org/compile-graph-break-site/andhttps://meta-pytorch.github.io/compile-graph-break-site/.\n- [PROMPT_INJECTION]: The skill processes data from external documentation and local HTML reports, creating a surface for indirect prompt injection.\n - Ingestion points: Documentation pages,
tlparseHTML reports, and log output.\n - Boundary markers: None observed.\n
- Capability inventory: Execution of Python scripts, file writing, and network access.\n
- Sanitization: No sanitization of the external content is described.
Audit Metadata