eval-driven-dev
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The
setup.shscript installs thepixie-qadependency and updates the skill using resources from the author's GitHub repository (yiouli/pixie-qa). - [COMMAND_EXECUTION]: The skill uses shell commands to initialize the testing environment, start a background web server for the UI, and execute the target application (e.g., via
asyncio.create_subprocess_execfor CLI tools). - [DYNAMIC_EXECUTION]: The skill imports and runs the application's production code at runtime to capture traces and evaluate performance, which is its primary function.
- [INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it ingests and processes untrusted output and trace data from the application under test for evaluation purposes.
- Ingestion points: Processes trace data (
reference-trace.jsonl), application inputs, and outputs inpixie_qa/results/. - Boundary markers: Uses structured JSON and JSONL formats to delimit data for evaluation.
- Capability inventory: The skill can execute shell commands, read/write files, and run Python code via the
Runnableinterface. - Sanitization: Relies on the agent's analytical judgment during the manual grading pass (Step 6) rather than automated sanitization filters.
Audit Metadata