skills/cursor/plugins/arena/Gen Agent Trust Hub

arena

Pass

Audited by Gen Agent Trust Hub on Nov 19, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill implements a multi-agent orchestration workflow where a primary agent reads outputs from multiple 'candidate' subagents end-to-end to synthesize a final artifact. This creates a surface for indirect prompt injection where a subagent could generate malicious instructions designed to influence the primary agent's selection, grafting, or verification process.
  • Ingestion points: SKILL.md Phase D ("Read every candidate end to end before picking") and Phase E ("Walk each losing candidate once more for what is worth porting into the base").
  • Boundary markers: Absent; the instructions do not specify using delimiters or ignore-instructions to isolate candidate outputs from the main context.
  • Capability inventory: SKILL.md identifies capabilities for spawning subagents (run_in_background: true), file system operations (creating git worktrees or directories in /tmp), and model orchestration.
  • Sanitization: Absent; the instructions for Phase E ("Fold each graft in by hand") rely on the agent's judgment without explicit sanitization or safety validation of the content being merged.
  • [COMMAND_EXECUTION]: The skill instructs the agent to orchestrate and manage multiple background processes for subagent execution.
  • Evidence: Phase B specifies spawning subagents in a single message using the run_in_background: true parameter, which involves task management and potentially parallel command execution.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 19, 2026, 11:11 AM