rfd4-proteina-autoresearch
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill defines a research strategy layer that coordinates existing protein design tools without introducing malicious patterns or unauthorized operations. It follows legitimate vendor-associated metadata conventions and includes security-conscious instructions for handling credentials.- [INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it ingests external data including design metrics and oracle predictions into a campaign database (campaign.sqlite).\n
- Ingestion points: Metric outputs and oracle predictions from rfd4-proteina-evaluate and rfd4-proteina-search are processed and recorded during design waves.\n
- Boundary markers: None specified in the instructions for isolating or delimiting external data from agent commands.\n
- Capability inventory: The skill uses Bash to execute subordinate tools and Write to record campaign outcomes and decisions to disk.\n
- Sanitization: No explicit sanitization or validation of the ingested scientific data is described, though the skill mandates keeping the campaign inspectable as a manual review mitigation.
Audit Metadata