worldos-simulation-review

Pass

Audited by Gen Agent Trust Hub on Aug 12, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill implements a restricted diagnostic workflow using WorldOS MCP tools for read-only access to simulation saves and turn history. It does not include any tools or instructions that allow for data modification or deletion.
  • [SAFE]: Robust negative constraints are explicitly defined to prevent the exposure of internal operations, file paths, credentials, and system prompts, protecting the underlying platform's security boundaries.
  • [INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it processes untrusted simulation prose. Ingestion points: Turn history records retrieved via get_owned_save_turns in SKILL.md. Boundary markers: Absent, as the agent reads prose directly from simulation logs. Capability inventory: Read-only MCP tools for inspection of save state and turns as defined in SKILL.md. Sanitization: Instructions require the agent to reconstruct behavior in product language and strictly forbid the exposure of internal raw operations or hidden prompts. This surface is assessed as safe due to the read-only nature of the tools and the specific diagnostic purpose.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 12, 2026, 02:46 PM
Security Audit — agent-trust-hub — worldos-simulation-review