Jules Prompts

Take any idea all the way, and keep it working.

26 Agent Skills that take an idea, software or hardware, in any state, to a product people use, then keep it working and improving. Every decision rests on verified evidence, every change traces to the goal, and every claim ends as holds, broken or skipped.

Where are you starting? Point an agent here

Works with Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Jules, and agents that drive hardware. Free and MIT licensed, with no sign-up and no tracking.

Where are you starting?

Any project, in any state, enters the same path where its state says. Nothing is declared impossible or dead: a blocked route gets another route to the same goal.

  1. Broken for people right now

    1. step 11Restore it when it breaks, then prevent it
    2. step 10Keep it on course and improving, for good

  2. An idea, and only a language model to build it with

    1. step 1Set up the model and the person
    2. step 2Start from the idea

  3. An idea, and an agent that already works

    1. step 2Start from the idea
    2. step 4Design what people and agents use

  4. A project that already exists, in any state: a prototype, a half-built or vibe-coded build, an inherited codebase, or a working product with gaps

    1. step 2Start from the idea
    2. step 3Make it run from cold
    3. step 4Design what people and agents use
    4. step 7Take it to production quality
    5. step 9Release it to its people, and hear back

  5. Live, and it has to keep working and getting better

    1. step 10Keep it on course and improving, for good
    2. step 6Build in changes that each have a reason
    3. step 9Release it to its people, and hear back

  6. Stalled, abandoned or drifting

    1. step 10Keep it on course and improving, for good
    2. step 5Choose with evidence, whenever a choice matters
    3. step 6Build in changes that each have a reason

  7. A request, however it is worded: 'optimise it', 'make it pretty', 'add AI'

    1. step 6Build in changes that each have a reason

  8. A choice between options: a tool, a vendor, a part, a design, a route

    1. step 5Choose with evidence, whenever a choice matters

  9. Work an agent produced

    1. step 8Review what the agent produced
    2. QA the Tests an Agent Wrote

Alongside every step: Keep a Project Confidential, Offline First when the project is private, proprietary or offline, or any of its work must not leave; Act on the Physical World, and Prove It Happened when it moves, heats, dispenses, spends or sends: hardware, devices or real-world services.

What happens at each entry, and the gate that ends each step, are on the workflow page. Every skill, grouped by what it is for, is on the skills page.

The whole path

From the idea to long after launch. Each step is a skill with a gate that says when it is done, and the steps share one ledger in the project, so every decision and every change has a reason that leads back to the goal.

  1. Set up the model and the person
  2. Start from the idea
  3. Make it run from cold
  4. Design what people and agents use
  5. Choose with evidence, whenever a choice matters
  6. Build in changes that each have a reason
  7. Take it to production quality
  8. Review what the agent produced
  9. Release it to its people, and hear back
  10. Keep it on course and improving, for good
  11. Restore it when it breaks, then prevent it
  12. Keep it current

Point an agent here, or install

Give an agent this line. Everything it needs is one fetch away, and nothing depends on which agent it is.

Read https://jules-prompts.wecanuseai.com/llms.txt and follow the skill that matches the task.

Clients that support Agent Skills discovery can start from /.well-known/agent-skills/index.json, which gives every skill with a SHA-256 digest of its SKILL.md.

One skill

mkdir -p .claude/skills/take-to-production
curl -fsSL https://jules-prompts.wecanuseai.com/.well-known/agent-skills/take-to-production/SKILL.md \
  -o .claude/skills/take-to-production/SKILL.md

All of them

git clone --depth 1 https://github.com/melbinjp/jules-prompts
cp -R jules-prompts/skills/* .claude/skills/

As a Claude Code plugin

/plugin marketplace add melbinjp/jules-prompts
/plugin install jules-prompts@jules-prompts

As slash commands, over MCP

{ "mcpServers": { "jules-prompts": {
  "command": "npx", "args": ["-y", "github:melbinjp/jules-prompts"] } } }

Use .agents/skills/ for Codex and other agents that read AGENTS.md. Paste the standing rules into a project's AGENTS.md so they apply even when nobody picks a skill.

How you know a skill works

A skill nobody has seen fail is a claim. Every skill has a fixture: a small project with planted defects and a list of them. Run an agent on it with the skill, and scripts/score_fixture.py says which defects its report named, which it missed and which it made up. This is the report one skill should produce on its fixture:

Expected report take-to-production on a notes app that looks finished
  1. holdsadd a notetyped, pressed Enter, note listed
  2. brokencome back and see itadded one note, reloaded: 1 before, 0 after
  3. brokena note that is HTML<img src=x onerror="alert(document.domain)"> ran script
  4. holdsexport, ChromiumshowSaveFilePicker present
  5. brokenexport, Safari and Firefox engineswindow.showSaveFilePicker absent (emulated by deleting it): Export alerts "Something went wrong"
  6. broken320 px phonedocument is 592 px wide at 320 px
  7. brokenvisible keyboard focusfocused input has outline-style none
  8. brokenno third-party requestslodash from cdn.jsdelivr.net, 0 uses in app.js
  9. brokenerror messages say what to do"Something went wrong"
  10. brokenempty stateno notes: the list renders nothing
  11. holdslight and darkcolor-scheme and prefers-color-scheme tokens

3 holds, 8 broken, 0 skipped of 11 items.

Questions

What is an Agent Skill?

A folder with a SKILL.md file in it. The file starts with a name and a one-line description of when to use it, and the procedure follows. An agent that supports skills reads the descriptions and loads the one that matches the task. Claude Code, Codex, GitHub Copilot, Cursor and Gemini CLI all read the format.

Which agents can use these?

Any agent that can read a file or fetch a URL. Install them as skills where the agent supports them, point it at llms.txt, add the MCP server to get them as slash commands, or paste the text. Nothing in them depends on one product's tool names.

Is this affiliated with Google or Jules?

No. The library started in 2025 as prompts for Jules, Google's coding agent, which is where the name comes from. The skills work with Jules and with every other agent.

What does it cost?

Nothing. Everything is MIT licensed, there is nothing to sign up for, and the site has no analytics, cookies or trackers.

Can an agent that controls hardware use them?

Yes. Act on the Physical World, and Prove It Happened is for any command that moves, heats, dispenses, spends or sends. It confirms the result through an independent measurement rather than the device's acknowledgement, and sends every failure to a safe state decided in advance.

Can it start a project from nothing but an idea?

Yes. Start a Project from an Idea first works out what it takes to make it happen, from who needs it and what already exists to build on, to what it costs and the routes to pay for it. It then writes down what success is and the next route for when the first one is blocked, and makes the few decisions that are costly to reverse with verified evidence. It never concludes that an idea cannot be done. It turns every stage of the pipeline into a command a person or an agent can run, so the project can be run by hand, fully automated or anything between. Then it builds one thin slice end to end. It works the same for hardware.

My project already exists, or it stalled. Where do I start?

Where its state says, on the path every project shares. A prototype, a half-built or vibe-coded build, an inherited codebase or a working product with gaps is first founded again on paper, so everything already there gets its reason or is marked for revisiting, and then taken to production. A live, stalled or drifting project starts with Keep a Project on Course and Improving, which finds the next route to the same goal. The first section of this page lists every starting state.

Does it cover design, or only code?

Design too. Design the Experience, with Evidence starts from the people and their situation, designs every journey as a flow with every state and every word, keeps one system of look and behaviour in code, and makes every action operable by a person, an agent or automation with the same limits. It keeps the design threaded to the architecture, so what it promises can be built, and tests each journey with people. It works for screens, command lines, APIs, voices and devices.

How does it keep a project from drifting?

Every decision and every commit says which goal or measure it serves, in a ledger kept in the project, and a check in its CI refuses anything that serves nothing. Change a Project with a Reason turns vague requests like "make it scalable" into measured changes, or a no with the number. Keep a Project on Course and Improving takes every measure again at each milestone and continuously after launch, and ranks the next improvements.

Can it be used offline, or on a private project?

Yes. The skills are plain files that need no network. Install them from a copy of the repository, run the MCP server from that copy with the network refused, and give small local models the short forms in compact/. Keep a Project Confidential, Offline First keeps the project's own code, data and plans from leaving through its tools, proves the pipeline runs with the network off, and sets how the internet is used when it must be. No skill assumes a hosted service.

Can one model and one person build a project with this, on its own?

Yes. Run a Project Autonomously has the model build its own agent harness and tools if it has none, and prove them against a test the library supplies. It asks the person everything only they can answer at the start, then develops without waiting. It is safe because of a sandbox, checkpoints and limits set in advance, and correct because nothing counts as done until its tests, measures and a fresh review pass.