editorial-room

Pass

Audited by Gen Agent Trust Hub on Aug 26, 2026

Risk Level: SAFE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided content, creating an attack surface where embedded instructions could attempt to influence sub-agents. \n
  • Ingestion points: SKILL.md (Step 1 and Step 2) reads draft text and attached artifacts (images, video, links) from the user context. \n
  • Boundary markers: None identified; the instructions do not specify delimiters or directions to ignore instructions embedded within the draft. \n
  • Capability inventory: The skill spawns multiple sub-agents (Step 3 and Step 5) and may use tools for factual verification (Step 2). \n
  • Sanitization: None identified; there is no mention of filtering or escaping user input. \n
  • Context: The adversarial design of the room (e.g., the skeptical reader role) and the final requirement for a primary agent decision serve as functional mitigations against simple indirect injections.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 26, 2026, 12:04 AM
Security Audit — agent-trust-hub — editorial-room