agentleFS
KazKato91KazKato91/roomi-claude-plugin/skills/deploy/SKILL.md

deploy

Skill0 starsChanged 21 days agoMITView on GitHub

Deploys the project in this repository to Roomi with the roomi CLI. It writes the guided story the room tours a viewer through, rehearses (the story's steps included), shows what would go live, and publishes only when the user says yes. Then it offers a viewer link, or on Free a Community listing. Use when the user says "deploy it to roomi", "deploy it to pitch-product", "put this on roomi", "ship this demo to roomi", asks for a pitch room or viewer link for the demo they built here, or asks for a tour or story for that demo.

  • Installs packages

162 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use pm-discovery, qa-phase and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • ww-w-aiSkill

    pm-discovery

    PM Agent Team - Automated product discovery, strategy, and PRD generation. Runs PM analysis workflow to produce comprehensive PRD before PDCA Plan phase.

    Also in 3 projects that use this

  • ww-w-aiSkill

    qa-phase

    Run QA tests at various levels (L1-L5) for features under PDCA management. Integrates with MCP tools to execute structured test plans.

    Also in 3 projects that use this

  • +10 more from ww-w-ai

The file

We split this skill into the rules it gives an agent

7 rules. 0 are common, 1 is distinctive, and 6 are very specific to this project.

  • 0

    Common

    Most of the 144 edited versions we read keep these.

  • 1

    Distinctive

    This author's own choices. Worth a look.

  • 6

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Find the project: Use the path given as an argument, or else the repository root of the current directory.

    line 35
  2. Very specific:

    Check the CLI and the sign-in: Run roomi --help. If roomi is missing, offer to install it, saying what the command does: it installs the …

    line 40
  3. Very specific:

    Run it only on a yes; on a no, give them the command and stop.

    line 47
  4. Very specific:

    Make sure there is a manifest: If there is no pitch.json at the project root, run roomi init there.

    line 52
  5. Very specific:

    Validate: Run roomi validate. It checks pitch.json and the story against the contract without building anything, including fields it does …

    line 97
  6. Very specific:

    Rehearse: Run roomi deploy (no --commit).

    line 116
  7. Very specific:

    Publish, on yes only: Run roomi deploy --commit and show its output, including the room URL it reports.

    line 134

“Common” is measured: kept by at least 6 in 10 of the 144 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Deploy to Roomi

Turn "deploy it to roomi" into a rehearsed deploy that publishes only on the user's yes. The roomi CLI does the work: it builds and runs the demo on this machine, which a remote server cannot do. Your job is to drive it, show what it says, and stop at each decision.

Rules that do not bend

  • Never run roomi deploy --commit without an explicit yes to the rehearsal you have just shown, in this conversation. "Deploy it" at the start is the request to rehearse. It is not a yes to publish. A yes to an earlier rehearsal does not carry over after anything changes. If you are unsure whether they said yes, ask.
  • Never edit the user's code or config on your own. roomi init writes pitch.json, pitch.README.md and PITCH.md, and adds one line pointing at PITCH.md to the end of AGENTS.md and CLAUDE.md (making AGENTS.md when neither exists). Those are the only changes made without a question; say which files it touched. Any other change, such as the single-origin change a multi-port repo needs, is shown as a diff and applied only on a yes.
  • Show the CLI's output, not a summary of it. The rehearsal's own words are the evidence the user decides on. Quote the part that says what would go live, in full.
  • Never print, echo or store a token or a viewer link's URL anywhere except the one reply that hands the link to the user.

Steps

  1. Find the project. Use the path given as an argument, or else the repository root of the current directory. If there is more than one likely project (a monorepo with several apps), list them and ask which one. Do not guess.
  2. Check the CLI and the sign-in. Run roomi --help. If roomi is missing, offer to install it, saying what the command does: it installs the Roomi CLI from npm, and needs Node 24 or later.

    npm install -g @getroomi/cli

    Run it only on a yes; on a no, give them the command and stop. Once it is installed, or if a later command says they are not signed in, ask them to run roomi login themselves (it opens a browser for them to approve) and stop until they say it is done. Never try to sign in for them.

  3. Make sure there is a manifest. If there is no pitch.json at the project root, run roomi init there. It detects the apps, their ports and routes, and writes pitch.json, pitch.README.md (what it found, and a story's shape on these apps) and PITCH.md. It never asks with nobody at the terminal: when it cannot tell something (usually an app's device), it stops, writes nothing, and names the flag that answers, such as --device app=laptop. Relay those questions to the user rather than answering them yourself, then run it again with their answers as flags. Show the pitch.json it wrote. Never pass --force unless the user asked to replace theirs.

    The type of room. roomi init also asks "A standard room, or an immersive one?", and with nobody at the terminal it writes a standard room. Relay it as the others: pass --room immersive, or add "room": "immersive" beside kind in pitch.json, only on the user's yes. Never choose it for them. The type changes only with a new deploy, and every viewer link stays.

    Then read PITCH.md. It is the guide roomi init writes for any coding agent, drawn from the docs of the roomi installed here: the rules, the container build contract, every field of pitch.json and how to write the story. Rely on it for what roomi checks and how; where it and this skill differ on a fact about roomi, it is the newer. The rules above, on when to stop and ask, stand either way. A project with a pitch.json and no PITCH.md, or one that names a command roomi refuses, gets roomi init --update: it rewrites PITCH.md and touches nothing else.

  4. If the repo serves on several ports (roomi init or roomi deploy will say so), a container build needs them behind one origin. Propose the change as a diff: what to add or change, and in which files. Explain it in a sentence and ask. Apply it only on a yes, and then go back to step 3. On a no, stop and say what is blocking the deploy.
  5. Write the story, if there is none. A viewer's room offers a guided tour built from story in pitch.json: a few steps, each saying what to notice, which app to show, and what to run on the demo first. It is how the demo gets told when the builder is not in the room, so a demo deployed without one is half-deployed. Draft it by reading the demo (its launcher, its demo or scenario routes, its seed data, its screens), following the story section of PITCH.md, or story.md, which has Columbus's story as a worked example. Show the draft as the story block you would add, and add it on a yes; it is a change to pitch.json like any other. If the manifest has controls.scenarios and no story, offer to turn them into one: scenarios still work, but the tour is the better way to tell it.
  6. Validate. Run roomi validate. It checks pitch.json and the story against the contract without building anything, including fields it does not know (a misspelt field would otherwise be silently ignored), and for a container build reads the Dockerfile against the build contract. Each problem prints the docs page that explains it; read that page before proposing a fix, and propose the fix as in step 5. Go on only when it passes.
  7. Look at it in the room. While the story is being written and fixed, roomi dev serves the real room on this machine around the demo's own dev server (start that first, as the project's README or package.json says; for a built static demo, roomi dev --static needs no dev server). It runs until stopped, so start it in the background, and give the user the room's address it prints. It redraws the room whenever pitch.json or a page changes, and prints what is wrong in roomi validate's and the rehearsal's words: a step's page that answers 404, a call the demo refused. Read those lines after each change and fix the story as in step 5. Nothing in it is recorded or sent, so it needs no yes. Stop it when you are done.
  8. Rehearse. Run roomi deploy (no --commit). It checks the manifest again, scans for secrets, builds, runs the build locally and checks that it answers, runs each story step's calls against that local copy and reports each step ok or FAIL, then shows what would go live. Nothing leaves the machine. Show the output. If it failed, show the failure and what the CLI suggests, and stop. A step that fails is usually a step that needs something set first (a clock, a login, an earlier step's data): propose the fix to the story, as in step 5. Do not try fixes in the user's code unasked.

    An immersive room's rehearsal adds a type: line. If it says FAIL type, the workspace's plan has no immersive rooms, or they are not available yet; nothing was built or sent. Show it, and ask whether to deploy it as a standard room (take the room line out) or move the workspace's plan first. Do not change room, or the plan, unasked.

  9. Ask. End your turn with one direct question, such as "Publish this to Roomi?", and wait. Anything other than a clear yes is a no.
  10. Publish, on yes only. Run roomi deploy --commit and show its output, including the room URL it reports. Run it as it is: never pipe roomi through tail or anything else, which hides its exit status.

If it fails, stop. Show its output in full, say what it says went wrong, and do not run it again: a deploy that failed once fails the same way, and each attempt spends one of the plan's deploys. A FAIL type after ok build: registered means the platform did not keep room: the build was not made live, so say so and stop. A fetch failed right after ok upload is not a network blip: the platform refused the build while unpacking it, usually for its size. Run roomi validate, and check the build against the static limits (at most 5,000 files, 32 MB per file and 256 MiB unpacked; a Next.js export writes every generateStaticParams entry as a page plus about five payload files), then propose the fix and rehearse again from step 8.

  1. Offer the next step for their plan. The deploy's closing lines say which. On Free there are no private links: offer roomi open to preview it, and offer to write a Community card and list it there, which is how anyone else sees it. On a yes, follow the roomi:list-on-community skill: it drafts the card from the story, rehearses it, and lists only on a yes. On a paid plan, ask whether they want a viewer link, and for whom (a name, and optionally an email and an expiry in days). If they give an email, ask whether Roomi should email the link to that person for them, or whether they will send it themselves; never assume either way. On a yes, create it with the roomi MCP server's create_link tool (send: true to email it: it rehearses first, then commit: true on their yes to what it shows; or send: false), or with roomi link <name> [--email <address> --send] if the MCP server is not connected. Give the URL to the user in that reply and write it nowhere else. Say that get_link (or roomi link copy <name>) shows it again if it is lost, that it can be revoked with revoke_link, and offer the Community as well.

Where the rules are written down

The docs are the reference for everything the CLI checks. Read the page before guessing at a rule:

Pages

The room's "Read more" pages are Markdown or HTML files. Those listed in pitch.json's pages upload with every deploy. To change them without deploying, roomi pages list, roomi pages add <file> [--title <title>] and roomi pages remove <id> act at once, so each needs the user's go-ahead first, as add_page does. roomi pages add refuses to replace a page of the same title unless given --replace: ask before passing it.

Afterwards

The same MCP server answers questions about the project once it is live: deploy_status, viewer_activity, read_feedback and pipeline change nothing. create_link, revoke_link and add_page each change something, and each needs the user's go-ahead in the conversation first. get_link shows a link's URL again and records that it did; use it when the user asks for one.

The server also has deploy_static, with deploy_guide and check_bundle beside it, for a static demo held in a chat with no terminal (claude.ai, the Desktop app); its instructions describe that case. Here, with the project on disk, use roomi deploy: it builds, runs and scans the project first, and deploy_static does none of that. That is the connector's own advice too: it suggests the CLI for a container demo, a build over its limits, or work in a repository with Claude Code, and offers to run the install where there is a terminal, as step 2 does.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.