Missions

Break large engineering work into planned, validated missions executed by specialized agents — each with clean context, tight scope, and independent validation.

Why This Exists

Single-agent sessions work well for narrow tasks, but real engineering work is often too large and messy for one context window. As an agent accumulates context, its focus and reliability degrade. Two specific failure modes:

• Irrelevant context buildup. Wide scope means the agent carries information that has nothing to do with its current sub-task. Signal-to-noise drops.
• Self-evaluation bias. An agent that just built something is a poor judge of whether it built it correctly. Its own reasoning biases it toward thinking what it did was right.