bootstrap
Pass
Audited by Gen Agent Trust Hub on Oct 25, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill specifies the installation of a Python package from a public registry to support its evaluation functions.
- Evidence: The instructions guide the user to perform
pip install py-openjudgeto access the grading and generation tools. - [DYNAMIC_EXECUTION]: The skill generates Python code snippets intended for execution that incorporate data from user interactions.
- Evidence: Steps 2 and 4 provide boilerplate code for rubric generators and grading runners that the agent is expected to populate and execute.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests raw product descriptions and examples which are used to configure the behavior of automated graders.
- Ingestion points: Step 1 Product Interview (SKILL.md) where user-provided system descriptions are collected.
- Boundary markers: Absent; the agent is instructed to summarize interview results into configuration objects without specific delimiters.
- Capability inventory: The skill requires Python execution and network connectivity to communicate with external LLM providers (e.g., OpenAI, DashScope) for grading tasks.
- Sanitization: The agent acts as a filtering layer by summarizing the raw interview responses into structured task descriptions and query samples.
Audit Metadata