Intent to spec
Each feature starts as the problem, who it's for and what done looks like. Claude drafts the spec from it and lists its concerns separately, so they're read before the spec is accepted.
New in 0.8: prompt review on every pull request
The project board for Claude Code
From idea to pull request, with Claude Code.
Describe the feature you want. Kumano turns it into a spec and a plan, has Claude Code build it with your sign-off at each stage, reviews the code before you do, and opens the pull request.
Free beta · Mac only, macOS 15.6+ · Apple silicon and Intel · Needs Claude Code
What is Kumano?
Kumano is a native macOS app that puts a project board and a product-management workflow on top of Claude Code. Each feature goes from a written intent to a spec, an approved plan, a build, an AI review of the diff and a pull request, and its intent, spec, plan and review are committed to your repository as Markdown files. It follows Anthropic's AI-Native SDLC Playbook and is made by Tall Poppy Studio.
- SwiftUI, native
- Local-first
- Runs your own Claude Code
- Free during the beta
How it works
One simple loop, from a one-line fix to a month-long feature.
01
Describe it
Tell Kumano what you want in plain words. It asks what problem you're solving and what "done" looks like, then puts the feature on your board.
02
Approve the plan
Claude Code studies your codebase and proposes a plan before touching a file. Approve it or ask for changes, and it builds step by step, checking in when it needs you.
03
Review and ship
See every change with the AI review beside it. Merge the pull request and press Release; the card moves to Done on its own.
Intent: Password reset by email
_Feature 8F3A1C2E… · captured · written by Kumano_
## Problem
People who forget their password have to email support and wait for a reply.
## Proposed outcome
A reset link arrives by email, works once, and expires after 30 minutes.
## Open questions
Should a reset sign the user out of other devices?
Spec: Password reset by email 2 concerns
_generated · written by Kumano from intent.md_
## Requirements
- Single-use token, 30-minute expiry, hashed at rest
- Same response whether or not the address exists
## Design
New ResetToken table; mailer reuses the sign-up template.
## Concerns
No rate limit on the reset endpoint yet.
Plan: Password reset by email approved
_approved · written by Kumano_
## Files that change
auth/reset.ts · db/migrations/0042_reset_tokens.sql · mail/templates/reset.html
## Order of work
1 migration → 2 token service → 3 endpoint → 4 email → 5 tests
## Proof
The project's own test command passes; a used token is rejected.
Review This Change
policy: REVIEW.md · findings never approve or block
ImportantToken compared with
auth/reset.ts:48
==; use a constant-time compare.auth/reset.ts:48
- if (token.hash == hash(input)) {
+ if (timingSafeEqual(token.hash, hash(input))) {
NitExpiry is a magic number; name it.
auth/reset.ts:12
auth/reset.ts:12
Nothing gets lost
Every decision is written down. You sign off at every stage.
Built on Anthropic's AI-Native SDLC Playbook. The intent, spec, plan and review for every feature are saved in your repo, so anyone on the team can see why a change was made.
Plan first, then build
Claude explores read-only and writes the plan. Edits start once you approve it, ask for changes, or deliberately skip the gate. Guided runs pause at each gate, and every status change records who made it.
Review before you read
When a feature is ready, an AI pass checks the diff against its intent, spec and plan using the review policy in your repo, and marks findings Important or Nit. A person still reviews the pull request.
Prompt review new in 0.8
Most wasted agent work starts with an unclear ask. Kumano scores the intent behind each session, explains the score and suggests a clearer rewrite. Opt-in, with your own API key.
Ship in one click
Kumano branches when work starts, commits each step and opens the pull request. The card moves to Done when the PR merges, and the Release button runs your release workflow.
Your rules, enforced
Built on Claude Code itself, so your CLAUDE.md, hooks and skills apply in every session Kumano runs. Governance lives in your repo as code, not in a settings page.
Prompt review
Code review checks what came out. Prompt review checks what went in.
When Kumano opens a pull request, it can review every Claude session behind it and post one comment, edited in place on re-run, plus a check on the commit. Teammates see how the change was asked for without installing anything.
- Scores the intent behind a feature from 1 to 5, explains why, suggests a rewrite
- Capture is local; the judge is off by default and uses your own Anthropic key, kept in the Keychain
- Formerly Promptly, now part of Kumano
promptly/prompt-review
For product and engineering
AI writes the code. Your team decides what gets built.
For product owners and founders
Founders and product owners describe the outcome in plain language. Kumano asks the questions a good engineer would and turns the answers into a spec Claude can build from. No terminal needed.
For developers
Developers get the depth of Claude Code without its sprawl: plans before edits, your hooks and skills honoured, a reviewed diff per feature, and a pull request that links back to the spec it came from.
Why Kumano
A real Mac app that keeps your work in your repo.
Native SwiftUI
A real Mac app, not a web view: menus, keyboard shortcuts, multiple windows.
Your repo is the database
Projects live on your Mac; intents, specs, plans and the review policy live in git beside the code.
Projects, not a session inbox
Project → feature → task, on a board. Sessions belong to the work that caused them.
Claude Code, done right
Depth in one agent rather than breadth across many. It runs the Claude Code you already use.
Latest releases
Shipping in the open.
Where Kumano is going
What's coming next.
- live in 0.8
Prompt review on every PR
One verdict per pull request, a comment plus a check, linking the intent, spec and plan it was built from.
- coming
Spend coaching
Reads your session history and suggests what to change: a premium model on a simple task, corrections that belong in
CLAUDE.md, a repeated workflow worth turning into a skill or hook. - coming
Cost per feature
Tokens, time and cost tracked against the card that caused them, rolled up per project.
- coming
Session replay
Step through a session turn by turn to see where it went sideways and which instruction would have prevented it.
- later
One board for the team
Intent reviewed the way code is: product owners write it, engineers sharpen it.
Questions
Frequently asked questions
What is Kumano?
Kumano is a native macOS app that organises Claude Code by your projects. Each feature moves from a written intent to a spec, an approved plan, a build, an AI review of the diff and a pull request; the intent, spec, plan and review are committed to your git repository as Markdown files. It is made by Tall Poppy Studio.
Does Kumano replace Claude Code?
No. Kumano runs the Claude Code CLI already installed on your Mac, so you need Claude Code installed and signed in. Because it is Claude Code underneath, your CLAUDE.md, hooks and skills apply in every session Kumano starts. Kumano never reads your Claude Code login.
Which Macs does Kumano run on? Is there an iOS app?
Kumano needs macOS 15.6 or later (macOS 26 is recommended) and runs natively on both Apple silicon and Intel Macs. There is no iPhone or iPad app.
How much does Kumano cost?
Kumano is free to download during the beta. Claude Code usage is billed by Anthropic under your own Claude plan or API key, the same as when you use Claude Code in a terminal.
Where does my data live?
On your Mac and in your repository. Projects are stored locally; each feature's intent.md, spec.md, plan.md and review are committed under .kumano/features/, while conversations, attachments and task files stay gitignored. Prompts go to Anthropic through Claude Code, as they would from a terminal.
What does "every stage leaves a record" mean?
Kumano follows the stages of Anthropic's AI-Native SDLC Playbook: intent, spec, plan, build, review and ship. The intent, spec, plan and review are written as Markdown files under .kumano/features/ and committed; build and ship leave commits and a pull request. A person signs off at each gate, so the reasoning behind a change sits next to the code in git.
What happened to Promptly?
Promptly, the prompt-analytics CLI, became Kumano's prompt-review engine in Kumano 0.8. The engine is still installed with npm i -g @getpromptly/cli, and the pull-request check keeps the name promptly/prompt-review. The product lives on as Kumano's prompt review.
Who makes Kumano?
Kumano is built by Akos Komuves at Tall Poppy Studio, an independent app studio in Hungary. Support and feedback: akos@tallpoppystudio.com.
Download Kumano
Free while in beta, with automatic updates. Questions or ideas? Email akos@tallpoppystudio.com.
Download for Mac- macOS 15.6 or later (26 recommended)
- Apple silicon or Intel
- Claude Code installed and signed in
- No iOS app
Prompt review · updated in place on re-run
8/10
## Intent clarity
The ask names the outcome and the edge cases. It doesn't say what should happen to pending invites when a member leaves.
## Built from
intent.md·spec.md·plan.md