building-screening-rubrics
Pass
Audited by Gen Agent Trust Hub on Jul 17, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill defines a methodical workflow for literature review tasks. It leverages platform-native research tools (rp_search, rp_abstracts, rp_cite, rp_fulltext) and explicitly instructs the agent to avoid insecure methods such as raw shell commands or unmanaged network requests.- [INDIRECT_PROMPT_INJECTION]: The skill processes external data in the form of academic abstracts and papers. While this is a vulnerability surface, the risk is addressed through the following architecture:
- Ingestion points: External abstracts and paper metadata are retrieved via the rp_search and rp_abstracts tools as described in Phase 2.
- Boundary markers: The skill utilizes structured JSON formats (abstracts-cache.json) to separate data, though it does not specify explicit LLM delimiters for abstract content within prompts.
- Capability inventory: The agent uses Python logic for scoring and file system tools to persist reviewed papers to papers-reviewed.json.
- Sanitization: No automated sanitization is performed on abstract text, but the mandatory 'Test-Driven Refinement' phase (Phase 2, Step 3) ensures that a human reviews and validates the agent's logic against ground truth before bulk execution.
Audit Metadata