Final Prompt

GPTClaudeGemini··▲ 381 copies·updated 2026-07-14

final-prompt-3.prompt

# Stage 3 — Final Synthesized Prompt

**Target file:** `case-study/03-advanced-feedback-loop/03-final-prompt.md`
**How it was produced:** The Architect's Draft v1 (`01-architect-draft.md`) was sent unmodified to four independent open-weight Critic models — `gpt-oss:20b` (OpenAI), `qwen3.6:27b` (Alibaba/Qwen), `gemma4:26b` (Google/Gemma), and `phi4-reasoning:14b` (Microsoft/Phi) — using `00-critique-request-template.md`. The four critiques plus the original draft were then handed to **Claude Sonnet 5**, acting as the synthesizing Architect, which reconciled the feedback into the final prompt below.
**Role boundary:** this is the *prompt*. It is executed by the fixed **code-generation model** (`gpt-oss:20b`) to produce `flappy_bird.py` in this folder — the same generator used for the zero-shot and standard stages.

> The block below is the exact final prompt, reproduced verbatim.

---

**PROMPT** (copy everything inside this block):

when to use it

Community prompt sourced from the open-source GitHub repo SinaHosseini7/multi-llm-feedback-loop (MIT). A "Final Prompt" style prompt — adapt the placeholders and specifics to your task. Imported as-is and not independently retested here, so check the output before relying on it.