debate

Pass

Audited by Gen Agent Trust Hub on Jun 21, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [SAFE]: No malicious patterns, obfuscation, or unauthorized access attempts were identified. The skill's operations are transparent and align with its described purpose of facilitating literary debate.
  • [PROMPT_INJECTION]: An indirect prompt injection surface is present because the skill ingests untrusted user draft content that is then provided as context to autonomous reviewer agents.
  • Ingestion points: Draft content and .status.yaml metadata files are loaded into the agent's context (SKILL.md, Step 1).
  • Boundary markers: The moderator logic uses markdown blockquotes and descriptive labels (e.g., "The passage:") to delimit user content within the deliberation challenges.
  • Capability inventory: The orchestration flow spawns and manages multiple concurrent AI reviewer agents (SKILL.md, Step 3 and Step 4).
  • Sanitization: No explicit filtering or sanitization of the input text is performed to prevent instructions embedded within user drafts from potentially influencing the reviewer personas.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 21, 2026, 11:51 AM
Security Audit — agent-trust-hub — debate