agent-exploration

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill acts as an orchestrator for researching external sources, which constitutes a surface for indirect prompt injection from malicious content found in the investigated files or web pages.
  • Ingestion points: The workflow involves sub-agents reading, searching, and scouting external sources and web content to identify relevant information as defined in SKILL.md and references/dispatch-rules.md.
  • Boundary markers: The skill uses a scoped-write contract and assigns specific questions to each worker to constrain the impact of any potential injection in the source material.
  • Capability inventory: sub-agents are restricted to read-only inspection commands and writing only to a single, named analysis artifact within a directory controlled by the parent agent, preventing broader system modification.
  • Sanitization: The instructions mandate that all substantive claims must be verified against cited evidence and that artifacts are inspected for validity and schema compliance before the final synthesis step (references/checklist.md).
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 04:17 PM
Security Audit — agent-trust-hub — agent-exploration