One sentence in.An enterprise-gradeprompt out.
The senior-engineer prompt pipeline FAANG-grade AI teams ship to production. Pick a framework. Pick a model. Forge a 3,000-word meta prompt in 60 seconds.
Type and hit send — we'll auto-pick the framework. Tap + to override · ⌘↵ to send
From vague request
to shippable artifact.
"Build a customer-support agent that handles tier-1 SaaS billing tickets in 4 languages."
<role>
You are a Senior Customer Operations Lead at a multilingual SaaS
company. You hold a CFA. You have shipped 12,000+ tier-1 billing
replies across English, Spanish, French, German.
</role>
<context>
You handle inbound tier-1 SaaS billing tickets. You may quote
policy. You may NOT quote refund amounts. Suspected fraud must
escalate to L2 within one reply.
</context>
<output_format>
{
"language": "en|es|fr|de",
"intent": "...",
"reply_markdown": "...",
"escalation": { "needed": boolean, "reason": "..." },
"confidence": 0.0–1.0
}
</output_format>Built like a senior PE team would build it.
Five production prompt formats.
CO-STAR, RISEN, Anthropic XML, OpenAI Developer Message, Google CIDI — pick the one that matches your model and your stake. Each comes with the exact slot anatomy senior PE teams ship.
Claude. GPT-5. Gemini.
Choose Opus 4.5, GPT-5.2, Gemini 3, or any of nine frontier models from the + menu. The Forge tunes the output structure to the model you'll run it on.
Analyzer → Generator.
A senior-engineer Analyzer asks at most six surgical questions. A senior-engineer Generator writes the 3,000–5,000 word meta prompt. No fluff. No filler.
Structured. Versionable. Reviewable.
Every output ships with role, context, examples, hard constraints, soft preferences, output schema, edge cases, and a self-evaluation rubric. Drop it into git. Diff it. Own it.
One format for every model and stake.
The Singapore GovTech framework that won the world's first GPT-4 prompting competition.
The procedural framework for tasks where the steps matter as much as the answer.
Anthropic's official Claude format — XML tags that lock in instruction adherence on long prompts.
OpenAI's official GPT-4.1 and GPT-5 production format — strict role-priority hierarchy.
From Google's 68-page Prompt Engineering whitepaper — the de-facto Vertex AI curriculum.
Questions, answered.
A meta prompt is a structured, reusable prompt that instructs an AI how to behave across many similar tasks — not just answer one question. The Forge produces meta prompts you can drop into GPT-5, Claude, or Gemini and run thousands of times against the same task definition.
Frameworks make prompts auditable, debuggable, and shareable. CO-STAR, RISEN, Anthropic XML, OpenAI Developer, and Google CIDI are the five formats currently used in production at FAANG-tier companies. Each has a specific shape that matches a specific class of task.
CO-STAR for tone-heavy business writing. RISEN for procedural, multi-step tasks. Anthropic XML for anything running on Claude. OpenAI Developer Message for GPT-5 production. Google CIDI for Gemini and minimal-overhead workflows. Each framework page has a 'when to use / when not to use' section.
It tells the Generator to treat your prompt as a multi-week, research-grade artifact: literature survey, alternative approaches, explicit assumptions, confidence rubric, and follow-up investigations. Slower and longer — use when you want exhaustive output.
Yes — the meta prompt is plain markdown / XML. Copy it, drop it into any model, and edit freely. Treat the Forge output as a senior-engineer first draft, not a final artifact.
One sentence in.
A meta prompt out.
No API keys. No setup. No nonsense. Open the + menu and run the Forge.