PipelineText → ImageMid creditEN

Multi-Agent Code-Generation Orchestrator

Planner / Coder / Tester / Reviewer agents coordinated by an orchestrator with shared memory.

When to use this prompt

For agentic-AI engineering papers describing structured code-generation or repair systems.

The prompt

A multi-agent code-generation orchestrator pipeline, drawn left-to-right with five specialized agents around a central Orchestrator.

Center: a circular Orchestrator block with the label "Orchestrator". A shared vector memory store (drawn as a small cylinder labeled "Vector Memory") and a sandboxed execution environment (drawn as a small box labeled "Sandbox") sit just above and below the Orchestrator.

Five agents arranged radially around the Orchestrator:
1. Planner — analyses the repository structure and generates a structured execution plan (plan.json).
2. Coder — produces a minimal unified-diff patch following the plan.
3. Tester — synthesises executable test cases for the proposed patch.
4. Reviewer — scores the patch on correctness, style and minimal-change criteria; returns structured feedback.
5. Repairer — applies Reviewer's feedback to revise the patch and re-runs tests.

Show typed message arrows between agents and the Orchestrator (plan, patch, test, feedback, revision). Loop arrow at the bottom labeled "iterate until tests pass".
Style: clean academic vector, navy / teal / amber palette, white background, suitable for ICSE / ASE / NeurIPS submissions.

Variations

Two-agent Critic-Reviser

Simplify to two agents: Critic and Reviser. The Critic flags issues with the current patch; the Reviser produces a revision. The loop terminates when no issues are flagged.

Tips

  • Place external resources (vector memory, sandbox) outside the agent ring to keep the loop readable.
  • Label each arrow with its message type (plan / patch / test / revision). Generic arrows obscure the protocol.
  • Use a centered Orchestrator with radial layout. Linear pipelines miss the iteration story.

FAQ

Try this prompt now

Open it inside the generator with the prompt pre-filled.

Try this prompt

Related prompts