chemeagle-guide
Pass
Audited by Gen Agent Trust Hub on Apr 5, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill defines a pipeline for extracting data from external, untrusted literature, which creates a potential surface for indirect prompt injection attacks.
- Ingestion points: Untrusted data enters the agent context via 'eagle.extract('paper.pdf')' and 'eagle.extract_batch(input_dir='...')' calls in SKILL.md.
- Boundary markers: No delimiters or 'ignore instructions' warnings are mentioned in the guide for isolating untrusted paper content from agent instructions.
- Capability inventory: The agent uses an LLM provider to process text and write results to a searchable database ('ReactionDatabase'), allowing external content to influence structured data or subsequent agent actions.
- Sanitization: The guide does not specify sanitization or validation steps for the content extracted from scientific papers before it is processed by the model.
Audit Metadata