optimizing-deep-learning-models
Pass
Audited by Gen Agent Trust Hub on Mar 12, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill definition in SKILL.md requests the Bash(cmd:*) tool, providing the agent with broad system command execution privileges.
- [PROMPT_INJECTION]: The skill has a potential surface for indirect prompt injection due to its processing of external data. 1. Ingestion points: scripts/analyze_model.py and scripts/optimize_model.py ingest model data and file structures from user-specified directories. 2. Boundary markers: Absent; there are no instructions to the agent to ignore potentially malicious content within the analyzed files. 3. Capability inventory: The agent is granted Bash(cmd:*), Write, Edit, and Glob permissions. 4. Sanitization: Absent; the provided scripts perform file operations and metadata analysis without sanitizing or validating the contents of the processed data.
Audit Metadata