screening-manager
Pass
Audited by Gen Agent Trust Hub on Mar 5, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external research paper files and protocol documents.
- Ingestion points:
output/PROTOCOL.md,papers/papers_raw.jsonl,papers/papers_dedup.jsonl, andpapers/core_set.csvin the local workspace. - Boundary markers: The workflow does not explicitly define delimiters or instructions to ignore embedded prompts in candidate data.
- Capability inventory: Operations are restricted to reading local files and writing to
papers/screening_log.csv. No network or subprocess capabilities are present. - Sanitization: No specific sanitization or validation of input file content is mentioned.
- Assessment: While an attacker could embed instructions in a paper title or abstract to influence the screening decision, the risk is low as the agent's capabilities are limited to producing a CSV file.
Audit Metadata