industrial-ai-research
Pass
Audited by Gen Agent Trust Hub on May 15, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted data from the web using
WebSearchandWebFetchtools during the literature collection phase. - Ingestion points: External research papers and metadata enter the agent context through the Phase 3 search workflow.
- Boundary markers: The skill uses explicit instructions to 'separate verified evidence from inference' and 'label preprints,' but lacks formal data delimiters to isolate external content from the main instruction flow.
- Capability inventory: The skill has access to local file utilities (
Read,Glob,Grep) and network tools (WebSearch,WebFetch). - Sanitization: Mitigation is present via a mandatory 'Verification and Triage' phase (Phase 4), a 'Contrarian Pass' in synthesis, and a 'Quality Gate' (Phase S4) that checks for placeholder leaks and consistent citation mapping before final delivery.
Audit Metadata