team-composition-patterns

Pass

Audited by Gen Agent Trust Hub on May 17, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [SAFE]: The skill is a collection of design patterns and heuristics for organizing AI agent teams, containing no executable code or malicious scripts.
  • [SAFE]: No remote code execution or unauthorized external downloads are present in the provided instructions or reference documents.
  • [PROMPT_INJECTION]: The Research Team preset configuration (found in preset-teams.md) creates a surface for indirect prompt injection through the use of web-based tools.
  • Ingestion points: Untrusted data enters the agent context via WebSearch and WebFetch tools as described in the Research Team members section.
  • Boundary markers: The provided task templates lack explicit delimiters or instructions for agents to disregard commands that might be embedded in retrieved web content.
  • Capability inventory: The agents are granted general-purpose status, which includes file system modification and shell execution (Bash) capabilities.
  • Sanitization: The instructions do not define or recommend sanitization steps for data retrieved from the web before it is processed by the agents.
  • [SAFE]: References to ~/.claude/settings.json in SKILL.md are for standard platform configuration of display modes (tmux, iTerm2) and do not represent a data exfiltration risk.
Audit Metadata
Risk Level
SAFE
Analyzed
May 17, 2026, 03:10 AM
Security Audit — agent-trust-hub — team-composition-patterns