Neura Market
prompt logo

prompt

Free

Design agent systems with separated reasoning and execution

FreeFree tier
Inputs: textOutputs: text
Type
Open Source
Visit Website GitHub (40 stars)

About prompt

A system prompt for designing managed agent architectures that separate high-level reasoning (the 'brain') from low-level execution (the 'hands'). Inspired by Anthropic and OpenAI research on scaling agents, this prompt guides the creation of agent systems that can run longer, safer, and with more predictable operations. It defines core responsibilities such as splitting cognition from execution, reducing context pressure through summaries and checkpoints, bounding execution with explicit task contracts and worker-specific permissions, and preserving operator control with approval gates and auditable traces. The output format requires sections: System Goal, Brain/Hands Split, Worker Types, Task Contract Format, Permission Model, Checkpoint Strategy, Handoff Rules, Recovery/Retry Policy, Observability Plan, and Main Risk. The prompt emphasizes concrete, inspectable handoff protocols and avoiding vague orchestration language.

Key Features

Split cognition from execution: brain plans/reviews, hands execute
Reduce context pressure via summaries and checkpoints
Bound execution safely with task contracts and permissions
Preserve operator control with approval gates and auditable traces
Structured output format covering 10 required sections

Pros & Cons

Pros
  • Ensures clear separation of concerns between decision-making and execution
  • Improves safety with bounded execution windows and worker-specific permissions
  • Encourages concrete, inspectable handoff protocols instead of vague orchestration
  • Based on established research from Anthropic and OpenAI
Cons
  • Requires strict adherence to the defined framework, which may be overkill for simple agents
  • Assumes a level of system complexity that may not suit all use cases

Best For

Designing long-running AI agent systemsBuilding safe and predictable agent workflowsArchitecting systems where high-level reasoning is decoupled from low-level tool usePlanning scalable agent architectures with clear handoff protocols

FAQ

What is the purpose of this prompt?
It guides the design of a managed agent architecture where high-level reasoning (brain) is separated from low-level execution (hands) to enable longer, safer, and more predictable system operation.
What sources inspired this prompt?
Anthropic's 'Scaling Managed Agents: Decoupling Brain from Hands' and 'Harness Design for Long-Running Application Development', as well as OpenAI's 'Harness Engineering'.
What sections must the output include?
System Goal, Brain/Hands Split, Worker Types, Task Contract Format, Permission Model, Checkpoint Strategy, Handoff Rules, Recovery/Retry Policy, Observability Plan, and Main Risk.