scienceclaw-reasoning

Pass

Audited by Gen Agent Trust Hub on Apr 6, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill involves processing untrusted premises and observations while having the capability to recommend or invoke tools, which establishes a surface for indirect prompt injection findings.
  • Ingestion points: The reasoning frameworks in SKILL.md process user-provided premises, observations, and experimental data.
  • Boundary markers: The skill provides structural templates (e.g., General Reasoning Template) but does not define explicit delimiters or instructions to ignore commands embedded in the data.
  • Capability inventory: SKILL.md identifies integration points with code-execution, scienceclaw-qa, literature-search, and scienceclaw-ie skills.
  • Sanitization: No explicit input validation or escaping mechanisms are described for the data processed during the reasoning steps.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 6, 2026, 11:31 PM
Security Audit — agent-trust-hub — scienceclaw-reasoning