Everything Claude Code needs, in one zip

What is in the zip

Five files are always there: CLAUDE.md, KICKOFF.md, docs/spec.md, docs/plan.md and docs/design-brief.md. Then one docs/flows/<name>.runic per ticked process and one docs/sequences/<name>.runic per ticked sequence diagram, each holding the text from your editor with a single trailing newline. File names are the document name lowercased with hyphens; two names that collapse to the same slug get -2 and -3, and a name with no letters or digits becomes untitled.

Runic writes the zip itself: files are stored uncompressed with a fixed timestamp, so the same files produce the same archive byte for byte. The download is named after the project, for example refund-desk-build-package.zip.

How the package is made

Click Build in the top bar. The Build Package window lists your processes with a checkbox and a line count each, plus a second list for sequence diagrams when you have any. Only documents with text appear. The project name is prefilled from the first process in your sidebar, ticked or not, and can be edited, up to 80 characters.

Start sends the ticked processes to the AI for a short interview. Generate package sends the same flows plus your answers, and the model returns one JSON object: overview, personas, user stories, data model, screens, integrations, non-functional requirements and open questions. Runic coerces it into a fixed shape, drops unknown keys, and renders every markdown file from templates. A non-JSON answer gets one retry with a stricter instruction; an answer with no goal, stories or screens counts as a failure too. Then Runic shows an error instead of a half-broken package.

What the spec contains

docs/spec.md opens with problem, goal, scope and out-of-scope, then a personas table with goals and permissions. Process Flows embeds each ticked process verbatim in a fenced runic block. User stories follow in the as-a / I-want / so-that form, each with Given/When/Then acceptance criteria rendered as a checklist. Then the data model, screens with key elements and states, integrations, non-functional requirements, Open Questions, and an appendix with every interview question and answer. An empty section prints None yet. instead of a gap. The headings are fixed English; the prompt tells the model to write the values in the language of your flows.

The plan and the design brief

docs/plan.md cuts the work into three ordered milestones. Milestone 1 is the data model entities, the first ticked process as the core flow, and the first third of the stories. Milestone 2 is the remaining flows, the screens and the second third. Milestone 3 is integrations, non-functional requirements and the last third. The split is contiguous on purpose; reorder in the file if your priorities differ.

docs/design-brief.md is the same spec seen from the screen side: a one-line product statement, the screen inventory with states, the flow order, the personas, and a pointer to docs/flows/ as the exact logic to design against. Copy design brief on the done screen puts it on your clipboard for Claude Design.

Limits and gotchas

Build needs a signed-in account with a verified email. The button only appears when you are signed in, so the /learn tour does not have it. The two AI stages share one limit of 12 calls per hour per account; a package that goes through first time uses two, and past the limit Runic shows Build limit reached — try again in an hour. Up to 40 processes are sent, each cut at 4,000 characters, 24,000 in total. Sequence diagrams are packaged as source but not sent to the AI, so the spec, plan and brief are written from processes only. Esc and a click outside close the window between stages, not while Runic is reading flows or writing the spec.

How to use it

1. 1
   Open Build

   Click Build in the top bar. The Build Package window opens with the project name prefilled from the first process in your sidebar.
2. 2
   Name the project

   Edit the Project name field if you like. It becomes the heading of every file and the name of the zip.
3. 3
   Tick the flows

   Under Processes in the package, tick the processes the spec should be written from. Sequence diagrams get their own list when you have any. The line underneath counts what is selected.
4. 4
   Start

   Click Start. Runic shows Reading your flows… and comes back with numbered questions.
5. 5
   Answer and generate

   Type what you know in the answer boxes, then click Generate package. Writing the specification… shows while the spec is produced.
6. 6
   Download

   The done screen says Package ready — N files and lists each path with its size in kB. Click Download .zip.
7. 7
   Unzip and hand off

   Unzip into an empty folder, run claude there, and paste the kickoff prompt. For Claude Design, paste the brief.

Commands & shortcuts

Tips

• Put the process everything else depends on first in the sidebar. The first ticked process becomes the core flow of Milestone 1, and the first process in the sidebar becomes the default project name.
• Untick flows you have not finished. Every ticked process is embedded verbatim in docs/spec.md, so a half-written flow lands in the spec as-is.
• Run ⚙ Audit + AI review on each process before you build. The spec inherits whatever gaps the flows have.
• Commit the unzipped files. Markdown and .runic are plain text, so a later package diffs cleanly against them in git.
• Name flows for humans. docs/flows/refund-request.runic is what Claude Code will read and quote back to you.

Questions