Jompter DocsAgent skills
The jompter-board and jompter-prompts skills — what they teach your agent, how to install them, and what to say to use them.
Jompter MCP gives an agent the tools. A skill teaches it how to use them well. Jompter ships two, and you install each with one click:
| Skill | For | Without it, an agent tends to… |
|---|---|---|
jompter-board | Drawing diagrams and wireframes on your boards | draw one shape per call, overlap labels, and never look at the result |
jompter-prompts | Working through a prompt you dictated | write to-dos without opening your screenshots, and mark things done that it never checked |
A skill is a short guide in a folder your agent reads. It adds no tools and no access: an agent with a skill can do nothing that the switches in the MCP panel do not already allow.
Installing them
- Click MCP in the top-right of the Jompter window.
- Scroll to Agent skills. There is a row per agent, each with Install Board and Install Prompts.
- Click the ones you want. Start a new conversation with your agent so it picks them up.
| Row | Installs into |
|---|---|
| Claude Code | ~/.claude/skills/ |
| Cursor | ~/.cursor/skills/ |
| Codex | ~/.agents/skills/ |
The skills change as the app does. When yours is behind, the button reads Update Board or Update Prompts; a tick means it is current. Nothing updates on its own. If a folder with the same name is already there and Jompter did not put it there, the button reads skipped and leaves it alone.
The skills only help if the matching switch is on: Let agents draw on your boards for the board skill, and …and work on your prompts as well for the prompts skill. Setup →
Using them
You do not have to name a skill. Your agent loads it by itself when what you ask matches —
mention a board, a diagram, a prompt id or your to-dos, and it reaches for the right one. In
Claude Code you can also call one directly: type /jompter-board or /jompter-prompts.
The board skill
Say what you want drawn, and on which board if it is not a new one. Copy a board's id from the chip at the top of the open board.
Draw our checkout flow as a sequence diagram on a new Jompter board.Sketch a signup screen wireframe on board 3f9c2a1e — email, password, a Google button.Tidy up the architecture diagram on board 3f9c2a1e: the arrows cross and the labels overlap.What the skill has the agent do:
- Find the right board first, and never draw on one it is not sure you meant.
- Plan the layout in words before drawing, so shapes do not land on top of each other.
- Use the built-in library for buttons, form fields, charts and icons rather than building them out of rectangles.
- Draw in small batches, in the order it would explain it, so you watch it take shape.
- Look at the result and fix overlaps and cramped labels before it tells you it is done, then frame the finished diagram on your screen.
- Save a snapshot before a big change, and roll back rather than pile fixes on a mess.
- Ask before clearing or deleting a board it did not create in that conversation.
The prompts skill
Dictate your prompt in Jompter — talk through the problem, take screenshots of what you mean. Then open it, click the id chip in its header, and paste the line into your agent:
Jompter prompt id: 4bc4c309-… (call get_prompt with it). Plan this.What the skill has the agent do:
- Read the whole prompt — the text, then every screenshot, and any recording's frames and transcript. What you said last often matters most, so it reads to the end before planning.
- Read your code so each to-do names the files it will change.
- Propose the list in one go and stop. It says what it would do first and waits.
- Revise what you send back. Click Changes on a to-do and write what is wrong; the agent rewrites it and resubmits it.
- Work only on what you approved, one to-do at a time, each marked doing while it is on it.
- Finish each one honestly — what it changed, what was really wrong, and how it checked: Verified in the running app, Tests only, or Not verified.
Things worth saying to it:
| You say | It does |
|---|---|
| "Plan this" | Reads the prompt and proposes to-dos, then waits |
| "Go" / "start on the approved ones" | Works through what you approved, in order |
| "What's left?" | Reads the list and tells you what is outstanding |
| "That one is not fixed — still happens on small screens" | Reopens it, reads its earlier attempt, and tries a different fix |
| "Carry on" (in a new session) | Picks up from the list and its notes, without redoing finished work |
The to-do list is the agent's memory. A new session, or a different agent, reads it to see what was already done and where the last one stopped — so you can close the conversation and come back tomorrow. How the list looks in the app →
When a skill does not seem to work
- The agent never mentions it. Start a new conversation — agents read their skills when a session starts. Check the button in Agent skills shows a tick for that agent.
- It says it cannot see your prompts. Prompt access is off. Turn on …and work on your prompts in the MCP panel, then reconnect the agent. →
- The button says skipped. Something else already uses that folder name. Move or rename that folder, then click install again.
- You use a different agent. The skills are plain folders with a
SKILL.mdinside — copy one from~/.claude/skills/into wherever your agent reads skills from.