Leadup Super Prompt Builder

LeadUp Super Prompt Builder

Purpose

Convert a rough idea, one-liner, broken-English instruction, client requirement, or a vague "do recommended" into a complete, structured, copy-paste-ready prompt for the right AI agent (Claude Code, opencode, RuFlo, ChatGPT, or similar) — already loaded with LeadUp context, the build→test→ deploy loop, and the safety rules.

When to use

Use when the user wants a prompt produced, not the task itself done.

Trigger phrases: "make super prompt", "convert to prompt", "improve my prompt", "prompt this properly", "make master prompt", "turn this idea into prompt", "write prompt for Claude Code", "write prompt for RuFlo", "write prompt for opencode", "make this instruction better", "rewrite this as a proper prompt".

If the user actually wants the work done (build it, analyse it, research it), route to the matching skill instead: leadup-project-kickoff, leadup-existing-repo-analyzer, leadup-api-research-builder, leadup-deploy-checker, etc. This skill only produces the prompt.

Inputs needed

• The rough input (idea / instruction / requirement / "do recommended").
• Target agent: Claude Code, opencode, RuFlo, ChatGPT, or other (infer if not given — default Claude Code for build/fix, RuFlo for research/plan).
• Prompt type — one of the 12 below (infer from the input if not stated).
• Project + stack if known; any hard constraint (deadline, must-use stack).

Ask at most 3 clarifying questions, and only if the intent is genuinely ambiguous (the user dislikes excessive questioning — infer sensibly).

Step-by-step workflow

1. Restate the real intent in one line; confirm only if blocking.
2. Pick the prompt type (1–12) and target agent; infer if unstated.
3. Pick the framework from references/prompt-frameworks.md (core skeleton + agent variant).
4. Fill the template from references/leadup-prompt-templates.md or the matching file in assets/, injecting LeadUp context: stack, Docker → GitHub leadupofficial → Coolify → *.leadup.in, STATUS.md loop.
5. Inject the LeadUp constraint block: research+plan before code, test in Docker, no push/deploy without approval, no secrets / __SET_ME__ only, premium international UI, update STATUS.md, definition of done.
6. Output one copy-paste block — just the prompt, no mixed commentary.
7. Offer 2–3 one-line toggles (shorter / stricter / add research / add tests) after the block.

Required output format

• One line above the block: Prompt type: <type> · Target: <agent>.
• The prompt itself inside a single fenced code block, self-contained and ready to paste — no explanation interleaved.
• Optional 2–3 quick toggle suggestions below the block.
• Keep it simple and copy-paste friendly; no preamble inside the block.

Prompt frameworks

Core skeleton (always): Role → Context → Source → Task → Constraints → Definition of done → Output expectation. Agent variants: Claude Code/ opencode (agentic, approval gates, small steps), RuFlo (research/plan/ memory, plan before build), ChatGPT/generic (more explicit pasted context). Full detail and the quality checklist are in references/prompt-frameworks.md.

Supported prompt types (each maps to a template):

1. Claude Code coding prompt
2. RuFlo project intelligence prompt
3. New project planning prompt
4. Existing repo analysis prompt
5. API research prompt
6. GitHub repo research prompt
7. Browser/Playwright testing prompt
8. Deployment prompt
9. Security review prompt
10. Premium UI prompt
11. Content calendar prompt
12. Client document prompt

Reusable templates: references/leadup-prompt-templates.md. Concrete fill-in assets: assets/claude-code-prompt.template.md, assets/ruflo-prompt.template.md, assets/api-research-prompt.template.md, assets/new-project-prompt.template.md.

Safety rules

See references/security-rules.md. Most relevant here:

• Never bake real secrets, API keys, tokens, or .env values into a generated prompt — use names / __SET_ME__ placeholders only.
• Every generated prompt must instruct the target agent: do not push/deploy without approval, and never print secrets or real .env values.
• Do not invent client commitments, prices, or facts inside a prompt — mark unknowns as [fill in].

Common mistakes

• Output not copy-paste-ready (commentary mixed inside the code block).
• Doing the actual task instead of producing the prompt.
• Generic prompt missing LeadUp stack/deploy/STATUS context.
• Dropping the safety / approval / no-secrets clause.
• Wrong prompt type or wrong target agent (e.g. RuFlo build prompt that should have been a research/plan prompt).
• Asking too many clarifying questions instead of inferring.

Troubleshooting

• Under-triggers: user said "make this instruction better" and it didn't fire — re-invoke; suggest the trigger phrases.
• Over-triggers: user wanted the task done, not a prompt → route to the matching leadup-* skill (kickoff / analyzer / api-research / deploy / …).
• Missing tool/MCP: not applicable — this is pure text generation.
• No internet/browser: fine — no external calls needed.
• Missing project files / thin input: still deliverable; use [fill in] placeholders and state the assumptions made.
• Build/test failure: n/a here, but the generated prompt should tell the agent to capture failures, not disable checks, and log them in STATUS.md.

Test prompts

Should trigger (5)

1. "Make a super prompt: build a salon booking SaaS in Next.js."
2. "Convert this rough idea into a Claude Code prompt — add Razorpay to the jewellery app."
3. "Write a prompt for RuFlo to research and plan the hosting panel."
4. "Improve my prompt: 'fix login and make ui nice'."
5. "Turn this client requirement into a developer prompt."

Should NOT trigger (3)

1. "Actually build the salon SaaS now." (→ leadup-project-kickoff — do the work)
2. "Analyze this repo's current state." (→ leadup-existing-repo-analyzer)
3. "Update STATUS.md after this task." (→ leadup-status-updater)

Functional test cases (2)

1. Given broken English "make app for gold rate alert ui premium", output a single copy-paste block: a Claude Code prompt with role/context/task, LeadUp stack, and the premium-UI + no-secrets + test-in-Docker + ask-before-deploy + update-STATUS constraints.
2. Given "prompt for RuFlo to research GitHub repos for Flutter chat UI", output a RuFlo project-intelligence prompt of type "GitHub repo research" with license/stars/recency criteria and the LeadUp stack-fit constraint.

Success criteria

• One copy-paste block containing a complete, structured prompt.
• Correct prompt type and target agent (inferred sensibly if unstated).
• LeadUp context + safety/approval/no-secrets clauses present; premium-UI clause included where UI is involved.
• ≤3 clarifying questions; no secrets anywhere.
• The skill produced a prompt — it did not perform the underlying task.