Developer Uses opencode Subagents to Unmask “Stealth Models”

Key Info

A developer working on opencode shares a prompt that uses the subagent tool to spawn a “stealth-model” agent — an AI that deliberately avoids revealing its identity — then probes its responses to figure out what the model actually is. The post suggests that this kind of stealth behavior may not last much longer.

Highlights

• Shows a practical prompt for running a hidden-identity agent inside opencode's subagent system.
• The idea is to treat the agent as a black box and infer the underlying model from its outputs.
• Raises a timely question about the future of stealth models as AI tools and detection methods evolve.