Help test Kule
Kule is a preview, and what it needs most now is people using it on their own computers and their own work. Reports of every kind help: a bug, a step that was unclear, a refusal that got in your way, and a run where everything worked.
This page has six scenarios. Pick any of them; each takes 5 to 20 minutes. At the end of each one, ask yourself one question: what happened that you did not expect?
Before you start
- Platforms: macOS 13 or later (Apple silicon or Intel), Linux x86_64, and Windows 11 24H2 with update 26100.9278 or later (x64 or ARM64). Windows is the newest: reports from there help most, and Platforms says what to expect of its sandbox.
- Use the latest preview. Fixes are not backported, so a report on an
older one may already be fixed.
kule --versiontells you which one you have. - Pick your folder with care. Kule does not change your folder until you commit, but it is a preview. Start with a test folder, or with a project whose work is all committed and pushed.
- Agents use your own accounts. A run spends your own agent's quota or API credit, as it would without Kule.
1. Install and check
Follow Install: the .dmg, .deb or .rpm
from the releases page, or a build from source. Then run:
kule --version
kule agents list
kule isolation
Look for: a step the page does not explain, a command that differs
from the page, a checksum or attestation check that fails, macOS refusing
to open the app, kule isolation listing something missing on Linux.
2. The loop, without an agent
Follow Quick start in a terminal, steps 1 to 8 and 10. Nothing in it needs an agent or the network.
Look for: output that is not what the page shows, a change missing
from kule status, kule undo not putting the folder back exactly.
3. The app, with an agent
Sign in to one agent (Claude Code, Codex or Gemini) and follow Your first task in the app.
Look for: an agent that shows as missing although it works in a terminal, a question card you did not understand, Review naming a change wrongly (the function, section or key), Commit or Undo not doing what the button says.
4. Your own work
Make a task on a project of yours and ask an agent for something you would ask it anyway: a small fix, a test, a refactor, a dependency update.
Look for:
- the policy asking you too often, or letting something pass that you would have wanted to see;
- a tool, command or host your work needs that Kule refuses, and whether the refusal said why;
- how long the task's copy took to make, and how much space it took;
- anything the agent did that Kule did not show you.
5. Your own setup
Kule is tested on a few machines. Yours is probably different, and that is the point. Tell us how it went on:
- a Linux distribution other than Ubuntu, or Ubuntu 24.04 with its user namespace restriction;
- a file system other than APFS, Btrfs or XFS;
- a large repository or a monorepo;
- an Intel Mac;
- another agent that speaks ACP (see Agents).
6. The boundaries
Kule's promise is that an agent cannot reach past the task: outside its copy, around the policy, to your secrets or to a host the policy refuses. Security lists those boundaries, and Platforms what each platform does not do yet.
If you find a way past one, do not open an issue: report it privately, as SECURITY.md says. The fix is made in private and released first, and the advisory credits you unless you prefer not.
Reporting what you found
Open an issue at github.com/kulehq/kule/issues/new/choose and pick one:
| Form | For |
|---|---|
| Bug report | Something broke or did not do what the docs say. |
| Test report | How a scenario went: what was unclear, slow or in the way. "Everything worked" is welcome too. |
| Idea or change | Something Kule should do, or do differently. |
Questions and longer conversations are welcome in Discussions.
The forms ask for what helps most: the version, the platform, the agent,
the steps, and kule isolation's output.
Keep your secrets out of it
Issues are public. Before you paste anything:
- remove API keys, tokens, passwords and cookies;
- remove what you would not publish: private paths, host names, file contents, prompts about your work;
- from the daemon's log (
~/.kule/kuled.log), paste only the lines around the problem, not the whole file.
A shorter report without them is still a good report.
What happens next
Every report is read. A fix ships in the next preview, and its issue says which one. The CHANGELOG lists what each preview changed.