delta-protein

Pass

Audited by Gen Agent Trust Hub on Oct 23, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill operates by invoking the delta-cli tool. It provides strict, mandatory instructions for JSON escaping and shell wrapping (e.g., using cmd /c on Windows) to prevent command injection from user-supplied parameters.
  • [INDIRECT_PROMPT_INJECTION]: The skill provides an attack surface for indirect prompt injection by retrieving data from external research databases.
  • Ingestion points: Untrusted content is ingested via PubMed literature searches, UniProt metadata queries, and PubChem compound searches.
  • Boundary markers: The skill defines a strict 'Evidence boundary' requiring the agent to only use business results from the delta-cli output.
  • Capability inventory: The skill is restricted to calling delta-cli for remote operations and using local libraries for file parsing; it lacks arbitrary execution capabilities.
  • Sanitization: The instructions require the agent to restate JSON keys in natural language and verify all URLs against the raw JSON response.
  • [EXTERNAL_DOWNLOADS]: The skill facilitates the download of protein structures and analysis reports from delta-bucket-public.yangtzeailab.com. These are legitimate vendor resources used for data exchange between the Science Server and the user.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 23, 2026, 02:19 AM