multi-agent-patterns-advanced

Pass

Audited by Gen Agent Trust Hub on Apr 7, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill implements a task decomposition pattern that is susceptible to indirect prompt injection. Ingestion points: The 'goal' parameter in the 'decomposeTasks' function within SKILL.md is directly interpolated into a system message for an LLM. Boundary markers: Absent in the demonstrated implementation, allowing user-provided goal strings to potentially influence agent behavior. Capability inventory: The pattern includes execution of shell commands (git worktree) and a dynamic capability registry for generic agent dispatch across the system. Sanitization: No sanitization or validation of the input goal is present in the examples provided.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 7, 2026, 10:29 AM
Security Audit — agent-trust-hub — multi-agent-patterns-advanced