Kane CLI Plugin for Claude Code

kane-qe is the Kane CLI plugin for Claude Code. While Claude works on your code, it shows what kane-cli is testing, the step each test is on, why a test failed, and how much of your requirements is proven. When Claude changes code that nothing has tested yet, it says so and offers a test. kane-qe only watches and offers. It never starts a test by itself: Claude runs kane-cli, because you asked or because you pressed a button that asks Claude.

kane-qe works alongside the Kane CLI skill. The skill teaches Claude how to run kane-cli; the plugin shows you what those runs are doing. Use both.

Before You Start

You need:

Claude Code with plugins that use mods (function-hook plugins), in the terminal or the desktop app.
Kane CLI, installed and logged in. See Installation and Authentication.
A project folder you open Claude Code in. kane-qe shows kane-cli runs whose folder is inside that project.

kane-cli runs use KaneAI credits. kane-qe itself spends none: the only kane-cli command it runs without being asked is the read-only coverage check.

Install

Inside Claude Code:

/plugin install kane-qe --marketplace LambdaTest/kane-claude-plugin

Or from a terminal:

claude plugin marketplace add LambdaTest/kane-claude-plugin
claude plugin install kane-qe@lambdatest

There is nothing to start after that. Open Claude Code in a project and the band appears above the prompt. For a whole team, commit this to the project’s .claude/settings.json. Everyone who opens the project and trusts it is offered the plugin:

{
  "extraKnownMarketplaces": {
    "lambdatest": {
      "source": { "source": "github", "repo": "LambdaTest/kane-claude-plugin" },
      "autoUpdate": true
    }
  },
  "enabledPlugins": {
    "kane-qe@lambdatest": true
  }
}

Your First Five Minutes

Open Claude Code in your project. The band shows idle and, on its third row, assurance coverage (or not set up).
Ask Claude to test something, for example: “Use kane-cli to check that a guest can add a product to the cart.”
Watch the band: the test’s name, its time, and the step it is on. The mascot’s eyes move while it runs.
In the chat, Claude’s kane-cli command becomes a one-line card with a spinner, then a card with the result.
Press View steps on the card, or type /kane, to see every step.

The Band

Three rows above the prompt, always there, beside the kane mascot. The first row always ends with [ Open ], which opens the pane. The third row is always assurance coverage.

running  checkout_guest 44s [ Open ]
⠹ Fill the shipping address
assurance ━━━━━━━━──────── 47% proven · 4 use cases

What is happening	Row 1	Row 2
Nothing has run in this session	`idle`	The last result in this project from an earlier session, such as `last run ✓ login · 2h ago`
One test running	`running`, its name and time	A spinner and the step it is on
Several tests running	`3 runs`, one coloured cell per run	`now` and each running test with its time
A suite running	`suite`, one cell per test, `5 of 12 tests done · 3 running · 4 left`	What is running now; a failure the moment it happens
A suite on the remote grid	`12 tests on the grid · sent 2m 05s ago`	The grid reports its tests only when the job ends
Runs finished	This session’s passed and failed counts	The newest failure and the step it failed on
Claude changed code, nothing tested it		`⚠ 2 files changed, untested` [ Test this change ]

A test started anywhere in the project shows up within two seconds: by Claude, in another terminal, or by a script. Runs from other folders do not appear. Every row fits the width of your terminal: names are shortened first, then items drop from the right.

The Card in the Chat

Every kane-cli test that Claude runs (kane-cli run, kane-cli testmd run, kane-cli testrun run) is drawn in the conversation as a card, in place of the plain shell row. While it runs, the card is one line:

⠹ kane · checkout_guest 32s · Fill the shipping address

When it ends, the card shows the result in three rows, with the kane mascot at its left edge:

╭───────────────────────────────────────────────────────╮
│kane · login_test.md      ✓ passed · 41s · 12.4 credits│
│7 steps                                                │
│[ View evidence ]                ⛨ Verified with KaneAI│
╰───────────────────────────────────────────────────────╯

╭───────────────────────────────────────────────────────╮
│kane · checkout_guest_test.md         ✗ failed · 1m 06s│
│on Fill the shipping address · Postcode field rejects …│
│[ View steps ] [ View evidence ] ⛨ Verified with KaneAI│
╰───────────────────────────────────────────────────────╯

The border is green for a pass and red for a failure.
A failed test shows the step it failed on and KaneAI’s own reason, from the run’s investigation. kane-qe never guesses it.
A suite shows one cell per test, its counts, and up to three failed tests. A suite where every test passed says ✓ all 3 passed.
View steps (or View tests for a suite) opens the pane on that run. View evidence opens the run’s evidence in your browser.
Claude Code folds several shell commands into one line (“ran 3 shell commands”). A folded line that holds one kane-cli test stays folded, with the card underneath. If it holds two or more tests, it unfolds so each test gets its own card.
Only tests that Claude runs in this conversation get a card. Other kane-cli commands, and other shell commands, keep their normal rows. A command that never became a test (for example, kane-cli was not logged in) keeps its normal row, so the CLI’s own error is readable.

The Pane

Open it with /kane or the band’s [ Open ]. It has three tabs. Runs. A card per run: failed first, then running, passed, waiting. Press a run’s name for its detail:

Each step with a tick or a cross and its time.
The step being worked on, or the one that failed, opens to what the agent thought, the action it took, and each check.
For a failure: why, kind (KaneAI’s category, severity and how sure it is) and where.
Open evidence.

A suite shows one row per test. A suite on the remote grid shows the grid, the job and its link. See Remote Runs. Assurance. Designed and proven coverage bars, how many tests are failing or blocked, when they last ran, and a card per use case. Press a use case to see what it still owes, under To design and To run, each with the reason and the command that does it. Close these gaps with Claude hands that list to Claude. History. The last 30 finished runs in this project, newest first: when each ran, how long it took, and where it failed or how many of a suite’s tests passed.

Testing What Claude Changed

When Claude edits code and no kane-cli test has run since, kane-qe notices. Edits to docs (.md, .txt and the like), hidden tool folders (.claude, .testmuai, .vscode) and git-ignored files do not count. What happens at the end of Claude’s turn depends on the After Claude changes code setting:

Setting	What happens
`offer` (default)	The band shows `⚠ 2 files changed, untested` with [ Test this change ]
`auto`	Claude is asked, once, to test the change before it finishes
`off`	Nothing

Test this change opens a card in the pane with:

The files that changed.
Your saved tests (in .testmuai/tests/) that mention the same feature, each with its last result and a Run it button. The closest matches come first; among equal matches, the test that failed last time leads. See Test.md.
If no saved test matches: Draft a new objective. Claude reads the change and writes a one-line test objective. Run the objective hands it to Claude to run.

The warning clears as soon as a kane-cli test that started after the edit finishes.

Assurance Coverage

If your project has a Kane CLI requirement store (a .context folder), the band’s third row shows coverage, such as 47% proven · 4 use cases, or 82% designed · nothing run yet before anything has run. kane-qe reads it with kane-cli cover gaps --json when the session starts and after each run. That command is read-only and spends no credits. See Assurance and Coverage. Without a requirement store, the row says not set up with [ Set up now ]. The Assurance tab then offers Set it up with Claude, which asks Claude to ingest your requirement (a PRD file, a Jira, Confluence or Linear link, or a web page), extract the use cases and design the tests.

Evidence

Kane CLI seals each run’s evidence into a pack when it exits. View evidence (on a card) and Open evidence (in the pane) start Kane CLI’s own evidence viewer on your machine and open it in your browser, in one click. A second press reuses the viewer that is already running. See Viewing Evidence. If no viewer can be started, Claude is asked to open the evidence instead. Where the chat cannot take clicks (see Limits), the card shows the evidence file’s path instead of buttons.

Commands and Settings

Command	What it does
`/kane`	Open the pane
`/kane assurance`	Open the Assurance tab
`/kane history`	Open the History tab
`/kane auto`	Claude tests a change before it finishes, for this session
`/kane ask`	The band offers a test after a change, for this session
`/kane off`	Nothing after a change, for this session

Settings are under /config → kane-qe:

Setting	Default	Notes
After Claude changes code	`offer`	`offer`, `auto` or `off`, as above
kane-cli command	`kane-cli`	How kane-qe starts Kane CLI, split on spaces. Change it if `kane-cli` is not on your `PATH`

Colours follow Claude Code’s theme: the design’s palette on dark themes, a darker one on light themes. A theme change shows within two seconds.

What It Runs and Sends

kane-qe makes no network requests of its own. It reads files on your machine, starts a few local programs, and hands text to Claude in your conversation.

Reads: Kane CLI’s live-run files under ~/.testmuai/kaneai/sessions/, your project’s .testmuai/ folder (saved tests, variables, evidence), and Claude Code’s theme setting. It keeps the last 30 results per project in its own store on your machine.
Starts:
- kill -0 and ps to check a run is alive and name it.
- tail to read new lines of a run’s events.
- git check-ignore and git diff for changes.
- kane-cli cover gaps --json for coverage.
- kane-cli evidence serve and your browser opener for evidence.
Hands to Claude:
- The changed files and their diff, when you press Draft a new objective.
- The prompt behind each button you press (Run it, Close these gaps with Claude, Set it up with Claude), shown in the conversation as your message.
- In auto mode, a request to test the change.

The plugin’s README lists every command and every prompt word for word. The privacy policy and terms of service apply.

Limits

Buttons need a surface that takes clicks: Claude Code’s fullscreen terminal or the desktop app. Elsewhere, cards show the evidence path as text instead of buttons, and the band and pane still update.
auto mode needs the end-of-turn hook. Where an organization manages Claude Code centrally, that hook does not run for plugins people install themselves, so Claude is never asked to test the change. Use offer there.
This project only. Runs started in other folders are not shown.
What Kane CLI does not report is left out: there are no step totals, no credits while a run is going, and no browser name.
A newer Kane CLI that changes its event format shows only each run’s name and time until kane-qe is updated, rather than showing something wrong. See Agent Output.

Troubleshooting

Problem	What to check
No band above the prompt	Run `/plugin` and check that kane-qe is installed and enabled. After installing, run `/reload-plugins` or start a new session
A test is not showing	Check that it runs inside the project folder Claude Code was opened in, and that Kane CLI is logged in (`kane-cli whoami`)
The third row says `not set up`	The project has no requirement store yet. Press [ Set up now ]
Coverage never appears	kane-qe could not run Kane CLI. Set kane-cli command in `/config` → kane-qe to the full path of `kane-cli`
Buttons do nothing	Use Claude Code’s fullscreen terminal or the desktop app
A card shows a plain shell row instead	Kane CLI did not start a test (for example, it was not logged in); the row shows the CLI’s own error

To report a problem, open an issue at github.com/LambdaTest/kane-claude-plugin.

Update and Uninstall

Update:

claude plugin update kane-qe@lambdatest

Then run /reload-plugins in any session that is already open. Uninstall:

claude plugin uninstall kane-qe@lambdatest

Next Steps

Skills for AI Agents: teach Claude Code, Codex CLI and Gemini CLI to run Kane CLI.
Agent Mode: the --agent output the plugin reads.
Evidence Packs: what every run captures and how to view it.
Assurance: requirement-linked test suites and the coverage on the band’s third row.
Source code and issues: github.com/LambdaTest/kane-claude-plugin

​Before You Start

​Install

​Your First Five Minutes

​The Band

​The Card in the Chat

​The Pane

​Testing What Claude Changed

​Assurance Coverage

​Evidence

​Commands and Settings

​What It Runs and Sends

​Limits

​Troubleshooting

​Update and Uninstall

​Next Steps

Before You Start

Install

Your First Five Minutes

The Band

The Card in the Chat

The Pane

Testing What Claude Changed

Assurance Coverage

Evidence

Commands and Settings

What It Runs and Sends

Limits

Troubleshooting

Update and Uninstall

Next Steps