- Home
- Skills
- Agents & Orchestration
- Staged Build Kit: Spec to Shipped Web or Desktop App
More screenshots
Works with the AI Agents you already use
Staged Build Kit: Spec to Shipped Web or Desktop App
Builds web and Tauri desktop apps from a requirements doc in approved stages.
$19
Staged Build Kit: Spec to Shipped Web or Desktop App
Example session with this skill installed
Build this with staged-web-lite
Requirements - Clay Corner weekend pottery class
A one-page site for my small pottery studio. People should see what the Saturday beginner class is, when it runs, what it costs, and sign up for a spot.
- Class: Saturdays 10:00-12:30, 8 seats, 45 dollars including clay and firing
- I want a sign-up form: name, email, which Saturday, any note
- I need to get the sign-ups somehow, I check email every day
- Photos: I have about ten photos of the studio and finished pieces
- I like warm, handmade-looking things. Not corporate.
- No budget for monthly fees if possible
- I have the domain claycorner.example
- Read your context and instructions
- Compiled the staged build kit
(The first reply, before any file is written.)
What I understood
You want a one-page site for Clay Corner, your pottery studio, at claycorner.example. A visitor should see what the Saturday beginner class is (10:00-12:30, 8 seats, 45 dollars including clay and firing) and sign up with name, email, which Saturday and an optional note. Sign-ups should reach you by email, which you check daily. The look is warm and handmade, not corporate, using about ten of your own photos, and it should cost nothing per month if possible. I have read the requirements file as it is and will not edit it.
Risks I see before we start
Overbooking. A simple form cannot count seats, so a ninth person can sign up for a full Saturday. → Treat each sign-up as a request that you confirm by email. → Your choice is question 2.
Personal data and spam. The form collects names and emails, and public forms attract bots. → One line saying what the details are used for, a hidden spam trap, and no sign-up data stored in the code repository.
Free form services have limits. Free plans usually cap submissions per month. → Fine at 8 seats a week; I will name the limit before you commit to a service. → Your choice is question 1.
Payment is not in the requirements. I am assuming people pay at the studio and the site only shows the price. → Tell me if you want online payment; it would make this a bigger project with fees.
- Ten photos can be slow on a phone. → I resize and compress them and load them as the visitor scrolls.
Questions - batch 1 of 2
Answer like "1-A, 2-B, 3-A, 4-other: ...".
1. How should sign-ups reach you? - this decides whether any outside service is needed
- A) A form service that emails you each sign-up (recommended) - free plan, no server to run, arrives in the inbox you already check
- B) Email plus a spreadsheet - each sign-up is also added as a row, so you can see every Saturday at a glance; a little more setup
- C) No form service - the button opens the visitor's own email app with the details filled in; nothing to maintain, but some visitors will not finish
- Other) write your own
2. What happens when a Saturday is full, and who keeps the dates current? - a plain page cannot count the 8 seats
- A) Sign-up is a request; you reply to confirm, and you edit one short list when a date is full (recommended) - simplest, free, fits 8 seats
- B) The page counts seats and closes a date at 8 automatically - needs a small database, more to build and to go wrong
- C) No dates on the page - visitors write the Saturday they prefer and you arrange it by email
- Other) write your own
3. Where should the site be hosted? - you have the domain; it needs somewhere to live
- A) A free static host connected to claycorner.example (recommended) - no monthly fee; I give you the exact steps
- B) Hosting you already pay for - give its name under "other"
- C) Decide later - I build it so it works on any host
- Other) write your own
4. Will the project's code repository be public? - planning notes and work logs sit in a docs/ folder next to the code
- A) Private repository (recommended)
- B) Code public, docs/ kept on your computer or in a private repository
- C) Everything public
- Other) write your own
Batch 2 will be the last: what must be on the page and what can wait, the look, and how we work together. After that I write a short plan for your approval. No code is written before you approve it.
Connects securely to your tools. The creator never sees your data.
About this skill
What it does
Give your agent a requirements file and it builds the app the way a careful contractor would: it asks before it assumes, plans in writing, gets your approval before each stage, and checks its own work before showing you.
Nothing runs ahead of you. Every stage starts with four lines - what, why now, deliverables, not included - and waits for a yes.
What makes it different
- It does not stop at the plan. Requirement and planning skills hand you a document. This kit carries the same project through build, review and a closing audit, with the plan as the thing every stage is checked against.
- Survives a dead session. The hand-off note is written before work is handed out, and every brief is a checklist the worker ticks as it goes. After a usage limit or a summarised conversation,
staged-resumecompares the ticks with git and tells you exactly where it stopped. Nothing depends on the conversation still being there. - Worker reports are not trusted. The main agent rereads the diff, reruns the checks and looks for deleted tests, disabled checks and out-of-scope changes before you see anything.
- Works with or without subagents. Six role cards (designer, ui, system, tester, helper, reviewer) are pasted into subagents when your agent has them. When it does not, the same briefs are carried out one role at a time.
- You are never asked a question you cannot answer. Every question comes as options with a recommended first choice and the reason. Design is chosen from real, clearly different screen variants, not from questions about typefaces.
- Nothing installed. No hooks, no config files, no agents in your home folder. Everything it writes is inside your project.
The four skills
| Skill | For |
|---|---|
staged-web-app | A full website or web app |
staged-web-lite | A landing page or single-feature tool, with lighter questions and documents |
staged-desktop-app | A Tauri 2 desktop app, from questions to a tested installer |
staged-resume | Picking up any of them in a new session or after a usage limit |
Use it when
- A one-shot "build me X" gave you something you did not ask for
- You want to see and approve the plan before any code is written
- The project is too long for one session and you keep losing context
- You are not a designer and want real variants to choose from instead of design questions you cannot answer
- You want a paper trail: decisions, briefs, a daily work log
How it works
- Questions - only what your requirements do not already answer, always as options with a recommended first choice. Risks you probably did not think of are raised first.
- Plan -
plan.md, numbered decisions, and a stageddev-plan.mdwith an owner and a done-when for every stage. No code before you approve it. - Stages - each one is introduced in four lines and approved. Work is handed out with a written brief to one of six roles: designer, ui, system, tester, helper, reviewer. If your agent can start subagents it uses them; if not, it works through the same briefs itself.
- Review - the main agent rereads the diff, reruns the checks and goes through the done-when list before you see anything. For screens, you look at the running app first.
- Records - a hand-off note is updated before every hand-out, so a session that dies mid-task can be resumed from files alone.
- Closing - a read-only audit for security, secrets, accessibility and performance before going public.
What you get in your project
docs/01-planning/- plan, decision record, design referencesdocs/02-development/- development plan with progress log, hand-off note, one brief per taskdocs/03-worklog/- what was done each daydocs/04-communication/- every question asked and what you answered- The app itself, built stage by stage
For desktop apps
staged-desktop-app adds what Tauri 2 projects get wrong late: who the app is for decides signing and updates up front; the app identifier is fixed in the first stage; permissions are opened one at a time with a reason; the installer is tested on a clean machine before release.
Where it stops
- You commit and push by default. It proposes commit messages; it can commit for you if you choose that at the start.
- You create signing keys, run signing and publish releases.
- It never force-pushes, rewrites history or makes a repository public.
- It is a process, not a template: it does not ship a starter codebase or lock you into a stack (desktop is Tauri 2).
What is in the package
Instructions and document templates only. No scripts, nothing installed on your machine.
- 4 skills
shared/workflow.md- approval gate, question format, records, git, secrets, resumingshared/delegation.md,shared/roles.md- briefs, six role cards, review checklistshared/questions-dev.md,shared/questions-design.md- what to ask and what the builder decidesshared/desktop/- Tauri 2 questions, stack rules, design differences, test layersshared/templates/- 10 project document templates
- It is slower than a one-shot build on purpose: it asks, plans and waits for approval.
- It does not ship a starter codebase. The stack is chosen in the plan (desktop apps use Tauri 2).
- Quality of delegated work depends on your agent. Without subagents the main agent does every role in turn, which takes longer.
- You run git pushes, signing and publishing. It gives the commands.
- It does not deploy for you. Hosting is decided in the plan and the deploy steps are given to you.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 4 days ago
Requires
Permissions
File Scopes
Reads and writes inside your project only: source code and a docs/ folder with the plan, briefs, hand-off note and work log. Runs the project's check, build and dev commands and read-only git commands. Uses a browser or web search, when available, for design references. Commits only if you choose that at the start and never pushes.