run-hugging-face-training-smoke
Pass
Audited by Gen Agent Trust Hub on Aug 26, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands for preflight checks and to launch the training runner script.\n- [PROMPT_INJECTION]: The skill features a surface for indirect prompt injection as it processes and executes script files (runner.py) which may contain instructions that influence agent behavior.\n
- Ingestion points: The runner.py file provided as an argument in the preflight check process in SKILL.md and scripts/preflight_runner.py.\n
- Boundary markers: The preflight command includes explicit --require and --forbid constraints to filter the input script's contents.\n
- Capability inventory: The skill possesses capabilities for shell command execution and local file system interactions.\n
- Sanitization: The scripts/preflight_runner.py utility uses the Python ast module to perform syntax validation and enforces string-based content restrictions on the target script before it is executed.
Audit Metadata