grounded-research

Pass

Audited by Gen Agent Trust Hub on Mar 28, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill consists entirely of markdown instructions aimed at improving the accuracy and reliability of the agent's research tasks. No malicious patterns, obfuscation, or unauthorized data access were detected.
  • [PROMPT_INJECTION]: The instructions are designed to constrain the agent's behavior to follow Anthropic's hallucination reduction guidelines. There are no attempts to override system prompts or bypass safety filters.
  • [INDIRECT_PROMPT_INJECTION]: The skill acknowledges that it will process untrusted external data such as web pages and documents, creating a surface for indirect prompt injection.
  • Ingestion points: External documents, web pages, and long-form sources (SKILL.md).
  • Boundary markers: The skill mandates the use of specific formatting for extractions (e.g., > "exact quote" — [Source]) to clearly distinguish source material from analysis.
  • Capability inventory: The skill utilizes the agent's existing capabilities for web searching and file reading.
  • Sanitization: It implements a defensive 'Extract Before Analysing' protocol, which forces the agent to explicitly identify source quotes before synthesis, acting as a manual verification step against injected instructions in external data.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 28, 2026, 09:01 PM
Security Audit — agent-trust-hub — grounded-research