testing
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructions and implementation guides (e.g.,
examples/test-context-implementation.md) direct the agent to ingest user-provided data models to generate significant portions of a testing framework. This creates an attack surface where a malicious schema or prose description could be used to influence the generated code or bypass AI safety constraints.\n - Ingestion points: The 'Data Model Discovery' phase in
examples/test-context-implementation.mdand the 'TestContext System' section inSKILL.md.\n - Boundary markers: No specific boundary markers or 'ignore' instructions are defined for the untrusted user input.\n
- Capability inventory: The agent generates multiple code files including generators, scenarios, and database adapters.\n
- Sanitization: The instructions require the agent to confirm the inferred model with the user, introducing a human-in-the-loop validation step.\n- [DYNAMIC_EXECUTION]: The TestContext implementation guide includes usage of browser-based dynamic code execution.\n
- Evidence:
examples/test-context-implementation.mdusespage.evaluate()to run thesignInWithCustomTokenfunction within a browser context for Playwright tests.
Audit Metadata