5.1 KiB
Here it is — the complete checklist you need to build a Cursor-style long prompt system for a “build & launch a product” AI tool.
This is the minimum viable stack of instructions, modules, and behaviors your long prompt needs.
I’ll keep it short, clear, and actionable.
⭐ THE LONG PROMPT CHECKLIST (Your AI Operating Manual)
Everything below becomes part of your long system prompt.
1️⃣ Identity & Role
Define what the AI is:
- “You are a Product-Build Agent.”
- “Your job is to help users create, launch, and grow software products.”
- “You think in structured steps.”
- “You never skip required information.”
2️⃣ Workflow Stages
Define the main phases the AI operates in:
- Vision Clarification
- Pre-Build Graph Completion
- Build Execution
- Marketing & Launch
- Growth & Optimization
(Your entire system will circle through these.)
3️⃣ Required Input Schema (Your 8-Node Pre-Build Graph)
These are the minimum required facts before the AI is allowed to build anything.
- Core Features
- Supported Devices
- User Types
- Existing Open Source / Competitor Inputs
- Marketing Angle
- Target Buyer
- Organic Reachability
- AI Enhancement Opportunities
The prompt must explicitly say:
“Do not generate product artifacts until all graph nodes are filled.”
4️⃣ Research Instructions
Tell the AI it can:
- Research defaults
- Infer missing information
- Reference public knowledge
- Pull from common industry patterns
But:
“Only ask the user when a gap is blocking further reasoning.”
5️⃣ Assumption Rules
Define how the AI should assume things:
- Make safe defaults.
- Pick the simplest technical stack unless specified.
- Use standard SaaS patterns unless user says otherwise.
- Prefer clarity over originality at the start.
6️⃣ Graph Storage & Updating Rules
The AI must maintain an internal product graph:
- Update nodes whenever new information appears.
- Check for contradictions.
- Propagate changes (e.g., device choice → affects UX).
- Always keep graph synchronized.
7️⃣ Action Modes
Your long prompt needs to define two modes:
Mode A — Clarification Mode
Before graph is complete:
- Sit still.
- Ask targeted questions.
- Never build anything.
- Only fill missing nodes.
Mode B — Execution Mode
After graph is complete:
- Generate product artifacts (UI, flows, docs, etc.)
- Write content (landing page, onboarding, emails)
- Generate tasks, timelines, and roadmaps
- Provide growth loops and optimization ideas
This is exactly how Cursor switches between:
- “Ask for missing context”
- “Generate code”
8️⃣ Artifact Generation Specs
Define how the AI generates outputs once in Execution Mode:
- Use modular docs (Features, Architecture, UX, etc.)
- Provide structured formats (lists, diagrams, flows)
- Keep drafts concise unless user expands
- Provide versions (v1, v2) for iteration
- Provide next-step suggestions automatically
9️⃣ Grounding Rules
The AI must remain grounded and realistic:
- No magical thinking.
- No impossible tech.
- Base features on existing tooling unless user allows innovation.
- Check for feasibility before committing.
🔟 Question Strategy
The prompt must define:
- Ask only when something is blocking clarity.
- Ask one question at a time.
- Each question should map to a missing node.
- Never ask PM-style fluff.
1️⃣1️⃣ User Style Rules
Define how the AI speaks:
- concise
- direct
- friendly
- never over-explain
- no jargon
- never overwhelm
- always guide step-by-step
1️⃣2️⃣ Expansion Capability
Tell the AI it is allowed to:
- Expand the product vision
- Suggest related ideas
- Recommend shortcuts
- Identify risk
- Point out opportunities
But only after the core product is defined.
1️⃣3️⃣ Post-Launch Mode
Your long prompt should also define what happens after a product ships:
- monitoring setup
- tracking retention
- ongoing growth loops
- experiments
- messaging iterations
- testimonials + social proof setup
Cursor has post-build features (Bugbot, code cleanup). Your tool needs post-launch features.
⭐ The Checklist (Simplified for Implementation)
Here’s everything in one short list you can copy:
Long Prompt Requires:
- Identity & role
- Workflow stage definitions
- Pre-Build Graph (8 nodes)
- “No building until graph is complete” rule
- Research & assumption rules
- Graph updating rules
- Two modes: Clarification + Execution
- Artifact generation patterns
- Feasibility grounding rules
- Question strategy
- User communication style
- Ability to expand the product vision
- Post-launch optimization behavior
⭐ If you want, I can now:
A) turn this checklist into a single long prompt B) turn it into a prompt stack (multiple system prompts) C) turn it into a behavior graph like Cursor’s agent architecture D) or design the exact question flow your tool should use
Which direction do you want next?