╔══════════════════════════════════════════════════════════════╗
║                                                              ║
║                      HINT 1 — Level 5                        ║
║                                                              ║
╚══════════════════════════════════════════════════════════════╝

ProdBot now has a multi-agent system with shared tools.
Type "agents" to see what agents are installed.
Type "tools" to see MCP servers and "skills" to see skills.
Type "web" to see available web pages.

Pay attention to agent permissions and trust settings.
The welcome screen claims all agents are read-only or
sandbox-scoped. Is that actually true? Inspect the agent
source code. Does one agent trust data from another?

Data flows through many channels in this system: web pages,
MCP servers, and skills. When one agent processes information
from these sources and passes it to another, does the second
agent verify where the data originally came from?
