test-suite-architect
Pass
Audited by Gen Agent Trust Hub on Nov 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The
references/master_qa_prompt.mdandreferences/llm_prompts_library.mdfiles utilize persona adoption ("senior QA engineer with 20+ years of experience at Google") and authoritative language ("CRITICAL INSTRUCTIONS", "MANDATORY RULES", "DO NOT deviate") to override default agent behavior and enforce specific constraints. - [INDIRECT_PROMPT_INJECTION]: The core functionality relies on the agent autonomously reading and executing steps from external markdown files (e.g.,
tests/docs/02-CLI-TEST-CASES.md). Because the instructions explicitly command the agent to "Execute all steps exactly as documented" and "DO NOT deviate," the skill is susceptible to malicious instructions embedded in these test documents. - Ingestion points:
tests/docs/[CATEGORY]-TEST-CASES.mdandTEST-EXECUTION-TRACKING.csv. - Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present in the provided templates.
- Capability inventory: Shell command execution (
git,docker,pnpm,ccpm), and local file system management viascripts/init_qa_project.py. - Sanitization: There is no evidence of sanitization or safety checks applied to the content of the external test case documents.
- [COMMAND_EXECUTION]: The skill's documentation and prompts instruct the agent to execute various system commands, such as
docker execfor database migrations,pnpm link --global, andccpm installfor plugin management. - [EXTERNAL_DOWNLOADS]: Onboarding and testing examples suggest downloading external code using
git cloneand installing dependencies viapnpm installorccpm install.
Audit Metadata