agent-advisor
Pass
Audited by Gen Agent Trust Hub on Nov 19, 2026
Risk Level: SAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- Dynamic Execution in Test Scripts: The skill uses
exec()andcompile()within its testing framework (specifically inscripts/test_opus55_generation_contracts.pyandscripts/test_runtime_default_docs.py) to execute code snippets extracted from its own Markdown documentation. This pattern is designed to ensure documentation accuracy by running unit tests against the examples provided to users. While these functions can execute code at runtime, they are applied here to internal templates rather than untrusted user inputs. - Potential for Indirect Prompt Injection: The discovery phase (
references/phases/discover/discover.md) ingests data by parsing user-supplied codebases to identify framework usage and model configurations. This introduces a surface where malicious files in a scanned repository could attempt to influence the agent's decision-making. The skill manages this by treating scan results as advisory hints and requiring explicit user confirmation for all detected signals before they are incorporated into the final architectural design.
Audit Metadata