threat-model-generation
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill analyzes untrusted repository content which could contain malicious instructions designed to influence the AI's security assessment.\n
- Ingestion points: The skill scans the codebase architecture and primary tech stack via files like package.json and requirements.txt (Step 1).\n
- Boundary markers: The instructions lack explicit delimitation to separate the code being analyzed from the agent's instructions.\n
- Capability inventory: The agent writes to the file system and executes shell commands for verification.\n
- Sanitization: There is no instruction to sanitize or validate content extracted from the repository before including it in the generated threat model.\n- [COMMAND_EXECUTION]: The skill uses standard shell utilities for verifying the success of its operations.\n
- Evidence: Use of
test,jq, andgrepwithin the Verification section to check file existence and content integrity.
Audit Metadata