skills/wakqasahmed/skills/roast/Gen Agent Trust Hub

roast

Pass

Audited by Gen Agent Trust Hub on Jul 16, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill operates within expected agent orchestration patterns for idea validation. No malicious code or configurations were detected. \n- [PROMPT_INJECTION]: The use of adversarial personas is well-defined and confined to the evaluation of user-provided ideas. No instructions to bypass safety filters were found. \n- [DATA_EXFILTRATION]: No unauthorized access to sensitive local files or environment variables. Subagents use web search for research, which is a legitimate capability. \n- [INDIRECT_PROMPT_INJECTION]: The skill has an indirect prompt injection surface. \n
  • Ingestion points: $ARGUMENTS and clarifying questions in SKILL.md. \n
  • Boundary markers: The 'THE BRIEF: [brief]' pattern is used for interpolation. \n
  • Capability inventory: Web search is available to subagents; no file system or command execution capabilities are present across scripts. \n
  • Sanitization: None present. \n The risk is minimal as the personas are adversarial by design, which naturally resists single-point instruction manipulation.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 16, 2026, 10:03 AM
Security Audit — agent-trust-hub — roast