AFP Prompt Architect｜Idea to Workflow
AFP = Auto-Flow Prompt, not the news agency. Describe your task in one line and get a copy-paste super prompt plus a .docx manual.
AFP Prompt Architect turns one vague line into a copy-paste super prompt. Diagnosis, SOP workflow, audited to V2.0, .docx manual included. Try it free.
Features
- Auto follow-up questions——Pull mode splits one line into scenarios, boundaries and red lines with at most one follow-up round — never a questionnaire
- Four modules, one build——Runtime protocol, multi-core engine, Phase-Step workflow and panoramic HUD, compiled per the AFP method, none missing
- Top-weight audit core——Executor and auditor cores red-team each other across completeness, consistency, executability, red lines and iterability before V2.0 ships
- Copy-paste plus archive——Full text in chat for one-click copy, plus a .docx manual with variable tables and version log — paste into any model
How it works
- Describe the task in one line Required: what you want AI to do stably. Optional: upload an SOP, old prompts or process docs (pdf/docx/txt/md, max 20MB). No idea? Tap a preset.
- Pick depth, language, delivery Auto-decide architecture depth for zero thinking; super prompt in Chinese, English or both; raw V1.0 or V2.0 after the five-dimension audit.
- Copy it, save the .docx The full super prompt appears in chat for one-click copy, plus a .docx manual with variable tables and version log — paste into Claude, GPT or Gemini.
FAQ
- Is AFP related to the news agency?
- Not at all. Here AFP stands for Auto-Flow Prompt: a methodology that turns one vague line into a runnable, auditable system prompt, first proposed by the Chinese AI community in the Gemini 3.0 era. The news agency AFP (Agence France-Presse) is something else entirely — this tool only architects prompts.
- How is an AFP prompt different?
- A normal prompt is a paragraph you must re-explain every time. An AFP super prompt is a reusable system: runtime protocol, multi-core engine, Phase-Step workflow and panoramic HUD, plus auto follow-ups and red-team review. Build it once, then only swap the topic.
- I only have one sentence. Enough?
- Yes — that is exactly its home turf. Type one line such as “ghostwrite long-form articles in my style” and diagnosis asks at most one follow-up round before orchestration starts. No ideas at all? Tap a preset (lesson prep, support SOP) and go.
- How do I use my SOP or old prompts?
- Upload a pdf, docx, txt or md file (max 20MB). The agent reads it first, then distills: reusable flows compressed into at most 7 phases, keepable prompt fragments kept, red lines recorded. Lesson notes, talk tracks and process docs all work — even phone photos of handwriting.
- Which AI models do super prompts work on?
- The generic build pastes into Claude, GPT, Gemini, DeepSeek and Kimi, and delivery includes cross-model usage notes. Honestly: system-prompt compatibility differs across models, so run complex jobs on your main model first, then migrate.
- How is super prompt quality checked?
- Through the built-in five-dimension audit: completeness, consistency, executability, boundary red lines and iterability. By default V1.0 is compiled first, then repaired into V2.0 with pressure-test cases and a version log. Pick “V1.0 direct” to skip it for small jobs.
- How long does it take? What does it cost?
- Delivered inside one conversation: the full super prompt plus a downloadable .docx (manual, variable tables, version log). Generate after login; credit cost follows the on-page display. Giant skills over ~8000 words ship in stages — V1.0 first, then iteration.
Try it free