Get Elite Prompts in 3 Modes
What is PromptForge?
PromptForge is a multi-agent prompt engineering workspace. Instead of one AI trying to do everything, PromptForge uses 18 specialized AI agents across 3 modes — each with a distinct role — to transform your rough ideas into production-ready prompts, structured plans, and visual diagrams.
The three modes:
- Visualize — See your idea as a diagram before you build it
- Plan — Get a step-by-step execution plan with risks and timelines
- Forge — Generate a production-ready prompt optimized for any LLM
The recommended flow:
Your rough idea
↓
Visualize (see it)
↓
Plan (structure it)
↓
Forge (prompt it)
↓
Paste into Claude/ChatGPT/Gemini → Elite output
The 6 Agents Behind Each Mode
Forge Mode
Plan Mode
Visualize Mode
5 Tips for Getting the Best Results
Tip 1: Be Specific in Your Input
The more specific your input, the better the output. Don't be afraid to add context.
❌ Weak input:
"Build me a landing page"
✅ Strong input:
"Build a landing page for my AI automation freelance services targeting SMB owners in Southeast Asia. I want to highlight 3 services: workflow automation, CRM integration, and invoice automation. Tone: professional but approachable."
The Scout agent will extract your intent — but it can only work with what you give it.
Tip 2: Use the Right Niche
Each niche tunes all 3 modes for your domain. Choosing the right one dramatically improves output quality.
Tip 3: Use the Full Chain for Complex Tasks
For simple prompts — Forge alone is enough. For complex projects — use the full chain:
Visualize → Plan → Forge
Why? Each mode carries context forward:
- Visualize clarifies the structure
- Plan adds phases, risks, and timeline
- Forge uses ALL of that context to write a prompt that's 10x more specific
The Forge output after the full chain is an execution prompt — paste it into any LLM and get the final deliverable immediately.
Tip 4: Use Optional Fields to Unlock Better Output
In Forge Mode, expand "Refine your prompt" to add:
- Target LLM — Claude gets XML tags, ChatGPT gets markdown headers, Gemini gets role anchoring
- Output type — system prompt vs one-shot vs chat starter
- Your context — "I'm a solo founder with no marketing budget" changes everything
These optional fields are processed by the Scout agent and change how ALL subsequent agents approach your prompt.
Tip 5: Use Test & Compare to See the Difference
Not sure if PromptForge is making a difference? Click "Test & Compare" after forging.
- Left panel: Your raw idea → generic LLM output
- Right panel: Your forged prompt → structured, expert output
The gap tells you exactly how much value PromptForge added. Use this to decide if you want to iterate further.
The Chain in Practice: Real Examples
Example 1: YouTube Channel Strategy (Content Niche)
Input: "I want to grow my YouTube channel about personal finance for Filipino millennials"
Visualize output: Content calendar flowchart showing pillar → repurpose → distribute loop
Plan output: 3-phase strategy (Foundation → Production → Growth) with 6-week timeline, parallel opportunities, and risks like "algorithm change midway"
Forge output: Complete YouTube content strategy prompt with Filipino-specific hooks, OFW audience considerations, Taglish tone guidance, platform-specific formatting rules, and KPI targets
Example 2: Portfolio Website (Code Niche)
Input: "Build an interactive portfolio with chatbot and fluid dynamics background"
Visualize output: Architecture diagram (Client → WebGL → Chat API → Backend)
Plan output: 4-phase development plan (Prototype → Integration → Testing → Deployment) with acceptance criteria and performance benchmarks
Forge output: Direct execution prompt with WebGL vs Canvas 2D decision tree, Navier-Stokes implementation guide, chatbot UI specs, auth requirements, and performance targets (60fps desktop / 30fps mobile)
Example 3: Medical Study (Medical Niche)
Input: "Explain sepsis pathophysiology to a 3rd year medical student"
Forge output (no Plan needed): 8-section clinical prompt with Sepsis-3 criteria, cytokine cascade, SOFA scoring, landmark trials (SMART, ADRENAL), clinical analogies, and qSOFA bedside pearl
Forking: When to Use It
Fork a project when you want to try a different angle without losing your original work.
Example:
- Original project: "Restaurant SaaS for small businesses in SEA"
- Fork at Plan: try "Restaurant SaaS for enterprise chains" with the same Visualize output
- Compare both Forge outputs → pick the better one
Fork depth: Maximum 3 levels (original → fork → fork of fork)
Exporting Your Work
After completing any mode, export your full project:
- Markdown — human-readable document with all 3 mode outputs formatted cleanly. Great for sharing with a team or saving to Notion.
- JSON — raw structured data. Great for developers integrating PromptForge outputs into their own workflows.
Export button is in the top-right of the workspace.
Understanding the Interrogator's Questions
After Forge Mode, the Interrogator agent generates 3 clarifying questions. These aren't filler — each question:
- Names a specific assumption baked into your forged prompt
- Explains what breaks if that assumption is wrong
- Helps you decide whether to refine your input and re-run
If a question surprises you — that's the Interrogator doing its job. Go back, add that context to your input, and re-forge.
FAQ
Q: Which mode should I start with? A: Depends on your niche. General recommendation:
- Clear idea, just need a prompt → Forge directly
- Complex project → Visualize → Plan → Forge
- Already have a plan → Plan → Forge
Q: Why does Plan Mode take longer than Forge? A: Plan Mode generates more structured output (phases, dependencies, risks, time estimates, Mermaid flowchart). The Plan Writer agent produces a full document — give it 45-90 seconds.
Q: Can I use the forged prompt on any LLM? A: Yes. Use the Target LLM selector in optional fields to optimize for your specific model. Without a selection, Forge produces a universal prompt that works on Claude, ChatGPT, Gemini, or any other LLM.
Q: What's the difference between Free and Pro?
Q: What is the Mermaid diagram in Plan Mode? A: Mermaid is a text-based diagramming language that renders as visual flowcharts. PromptForge generates and renders these automatically — you don't need to know Mermaid syntax. The diagram shows your project phases, dependencies, and critical path visually.
Q: Can I re-run a mode? A: Pro users can re-run each mode up to 3 times per project (revisions). Free users get 1 run per mode.
Pro Tips from the Builder
-
The Interrogator is underrated. Most users ignore the 3 questions at the end of Forge. Don't. Answer them, add the context to your input, re-forge. The second run is almost always significantly better.
-
Plan Mode before complex Forge runs. If your idea has multiple components, always run Plan first. The Forge output after a Plan is an execution prompt — it skips planning language entirely and produces a direct deliverable prompt.
-
Test & Compare is your quality check. Before using a forged prompt in production, run Test & Compare. If the gap isn't dramatic, your input might be too vague. Sharpen it and re-forge.
-
Use niche crews. General niche works for everything but excels at nothing. Switching to the right niche (Code, Marketing, Medical, Content, Business) changes the Wordsmith, Plan Writer, and Visual Prompter system prompts — the difference in output quality is significant.
-
Fork when exploring. Don't overwrite a good project trying to make it better. Fork at the stage you want to vary, experiment in the fork, keep the original safe.
PromptForge — Visualize. Plan. Forge.