Claude Code mod · kane-cli v16

kane-qe Field Guide

One /kane cockpit for KaneAI browser testing inside Claude Code: run tests in plain English, watch them live, find out who owns a failure from the evidence, run suites on the cloud grid and devices, and turn requirements into proven coverage.

KaneAI v0.8.20● default/prodTMS: KaneAI Generated / Untitled [ Run ] [ Insights ] [ Tests ] [ Assurance ] [ History ] [ Setup ]
Walkthrough

Every feature, recorded in a real session

    Recorded unedited in Claude Code with the mod loaded: real kane-cli runs against the LambdaTest ecommerce playground. Pick a chapter to jump to it.

    At a glance

    Where the mod shows up

    /kane …

    One entry point

    Type what you want: an objective runs it, a _test.md replays it, a word opens a tab, a question goes to Claude.

    side pane

    The cockpit

    Six tabs: Run, Insights, Tests, Assurance, History, Setup. Mouse, keyboard and hotkeys all work.

    above the prompt

    Live band

    While runs are in progress: the newest step, ◐ Kane ×2 for parallel runs, Open and Stop.

    status line

    Always current

    kane ◐ 2 running · … while running, then kane ✓ passed 2m ago; clears after 10 minutes.

    transcript

    Run cards

    When Claude calls a kane tool, its card streams steps live and offers Insights, Triage and CI.

    model tools

    Claude drives kane

    Nine tools: run, replay, suites, devices, assurance, tests, status, insights, generate.

    ~/.testmuai/kaneai

    Runs from anywhere

    Runs started in a terminal, CI or another agent are mirrored live and join History.

    .testmuai · .context

    Watches the project

    New tests, evidence and requirement commits refresh Tests and Assurance on their own.

    Flows

    How the work moves

    Run an objective

    flowchart TD
  A["/kane url objective
    or Run in the pane
    or Claude: kane_run"] --> B{"Objective coach"}
  B -- "no assertion or unset vars" --> C["Held: Let Claude add checks
    Run anyway · Edit"]
  B -- ok --> D["kane-cli run --agent
    steps stream live"]
  C --> D
  D --> E{"Verdict"}
  E -- passed --> F["Insights: clean or page signals
    Save as test · Playwright code · CI"]
  E -- failed --> G["Insights: who owns it"]
  G --> H["Fix the test with Claude"]
  G --> I["Draft bug report with Claude"]
  G --> J["Rerun · Check flakiness ×3"]

    Who owns a failure

    flowchart TD
  S["Failed run + evidence pack"] --> Q{"Out of credits or
    never started?"}
  Q -- yes --> N["DID NOT RUN: fix setup / top up"]
  Q -- no --> P{"New first-party errors
    at the failing steps?"}
  P -- "yes + kane says product" --> PB["PRODUCT BUG"]
  P -- "yes + kane says automation" --> NL["NEEDS A LOOK"]
  P -- "no (noise ignored)" --> K{"kane's category"}
  K -- "test data · locator · expectation" --> TW["The test itself is wrong"]
  K -- "step limit · timeout" --> RS["The run's settings stopped it"]
  K -- other --> CM["CLI loop or the model"]

    Requirements to proven coverage

    flowchart LR
  I["Ingest
    PRD · Jira · Confluence
    Linear · web page"] --> X["Extract
    use cases"]
  X -- "questions" --> Q["Answer by button
    or Let Claude answer"]
  Q --> X
  X --> R["Review
    Approve · Review with Claude"]
  R --> D["Design tests
    ACs → scenarios → _test.md"]
  D --> V["Fill variables"]
  V --> RT["Run designed tests
    testrun --from-context"]
  RT --> C["Coverage
    designed % · proven %"]
  C -- "requirement changed" --> E["Evolve stale"]
  E --> D

    Explore a web app

    flowchart LR
  A["Run scenarios
    pane · Claude · terminal"] --> L["Live strip
    every run in progress"]
  L --> H["History → Sites
    scenarios · pass counts
    pages reached"]
  H --> W["What to explore next
    Claude proposes 5"]
  W --> A
  H --> T["Triage the failures"]

    Suites on the grid and devices

    flowchart LR
  T["Tests tab"] --> W{"Run on"}
  W -- "This machine" --> L["Local Chrome
    or emulator / simulator
    doctor checks tooling"]
  W -- "Cloud grid" --> G["HyperExecute
    needs remote-execution plugin
    + MultiOS plan"]
  G --> D["Pick device: filter
    iPhone 15 · 17.5"]
  L --> R["Suite: per-test live steps"]
  D --> R
  R --> J["Job link · evidence · History"]
    The cockpit

    Tab by tab

    Run

    hotkey 1
    • Objective and URL fields; Run (r, visible browser) or Run headless (h).
    • The objective coach holds runs that verify nothing or use unset {{variables}}: Let Claude add checks, Run anyway, Edit.
    • Live steps, extracted values, warnings (low credits, unresolved variables), Test Manager link.
    • When several runs are in progress, a strip lists them all; click one to watch it.
    • After a run: Insights (i), Triage with Claude (t), Playwright code (p), Copy CI command (c), Ask Claude to save as test, Stop (s).
    • First-run guidance: install → login gates, then a checklist (installed, logged in, project, first run, saved test).

    Insights

    hotkey 2 · views 6 7 8
    • Reads the run's evidence pack: per-step model vs browser time, network (HAR) and console per step, kane's verdict and suggested fix.
    • Owner box: PRODUCT BUG, NEEDS A LOOK, CLI / MODEL, PAGE SIGNALS, CLEAN, DID NOT RUN. Third-party ad/tracker errors and errors the page already logged earlier are ignored as noise.
    • Wasted effort: unchanged pages, the same action twice (even worded differently), repeated extractions, step budget used up, long waits.
    • Timeline (6) waterfall: hover or click a step for its timing, requests, console and screenshot (drawn in the terminal). Signals (7) heat grid. Time split (8).
    • Follows the newest run; once you pick a run it stays, with Follow latest.
    • Evidence viewer (v) opens the pack on evidence.lambdatest.com; Validate pack; Rerun; Check flakiness ×3; Fix the test / Draft bug report with Claude.

    Tests

    hotkey 3
    • Saved *_test.md library: ▶ run, status (Test Manager identity), sync (push the bundle).
    • Run on: This machine or Cloud grid (HyperExecute). Target: Web, Android, iOS with a filterable device and OS picker.
    • Run all, Dry run (grid preflight costs nothing), run by tags. Suites show each test's live step.
    • Install grid plugin from here; a refused grid job shows HyperExecute's reason (plan, quota).

    Assurance

    hotkey a
    • Stage strip Ingest → Extract → Review → Design → Run & cover, with counts, designed % and proven % (and when it was last proven).
    • Ingest a file or a Jira, Confluence, Linear or web URL. Questions kane pauses on appear with option buttons (★ recommended) or Let Claude answer.
    • Per use case: Approve, Design tests, gaps with kane's next command. Run N designed tests, Fill variables with Claude, Review with Claude, Evolve stale, View graph, Sync with team.
    • Refreshes by itself when anyone (terminal, CI, teammate) commits to the store or runs a test.

    History

    hotkey 4 · views 6–0
    • Runs (6): every run with owner mark, insights, ↻ rerun, triage; QA report with Claude.
    • Trends (7): duration bars by result, failures by owner, flaky objectives. Credits (8): balance meter, runs left at your average, spend per run.
    • Quality (9): web vitals trend and Measure web vitals; requirement coverage.
    • Sites (0): exploration grouped by site with scenarios, pass counts and pages reached; What to explore next and Triage the failures.
    • Includes runs from terminals and CI on this machine, found on disk.

    Setup

    hotkey 5
    • Version, login, credits (shows an overdrawn balance plainly), project and folder, default URL, window.
    • One-click run settings: mode, assertion mode, bug detection, final validation.
    • Mobile and cloud grid: Check Android / iOS tooling (each failing check with its fix), grid plugin check and install.
    Reading a result

    Owner labels in Insights

    PRODUCT BUGkane says product, and the page logged new first-party errors where it failed.
    NEEDS A LOOKkane and the page disagree; check the screenshot before filing.
    CLI / MODELNothing new went wrong on the page. Headline names a wrong test, the run's settings, or the agent.
    PAGE SIGNALSPassed, but the site logged its own errors (a 5xx, a /undefined/ URL).
    CLEANPassed; only third-party scripts (ads, trackers) logged errors, if any.
    DID NOT RUNRefused to start or out of credits: fix setup or top up.
    Reference

    Commands

    TypeWhat happens
    /kaneOpen the cockpit
    /kane <objective>Run it now (“run” in front is optional); the coach may hold it first
    /kane https://site.com <objective>Run it on that URL
    /kane path/name_test.mdReplay a saved test
    /kane insights [words]Evidence of the latest run, or the newest run matching the words
    /kane tests | history | setup | assureOpen that tab
    /kane suite --tags smoke --parallel 3Run a suite locally
    /kane grid --tags smoke --parallel 4Run a suite on the HyperExecute grid
    /kane ingest <prd.md | URL>Start the assurance loop from a requirement
    /kane triageClaude diagnoses the last failure from its evidence
    /kane cases <feature>Generate test cases with KaneAI, reviewed by Claude
    /kane stopStop running tests and assurance commands
    /kane <question>?Ask Claude about your kane runs

    Tools Claude can call

    ToolDoesNotable inputs
    kane_runRun a plain-English objective and return the verdict, steps and extracted valuesurl, headless, max_steps, variables (secret), name (save as test), target / device_name / os_version / app (mobile)
    kane_test_runReplay a saved _test.md with self-healingpath, author, adaptive_heal
    kane_suite_runRun many tests as one suitetags, match, parallel, dry_run, remote (grid), device_name, os_version
    kane_devicesList emulators / simulators here or on the grid, with doctor resultstarget, remote
    kane_assureDrive the requirement loop headlessaction: status · ingest · extract · design · answer · approve · evolve · sync
    kane_insightsRead a run's evidence and say who owns the resultsession_dir (default: latest)
    kane_tests_listList saved tests
    kane_statusVersion, login, credits, project, settings
    kane_generateGenerate scenarios and cases; save them as testsobjective, limits, save + request_id

    Keys in the pane

    1–5, aRun, Insights, Tests, History, Setup; Assurance
    6–0The open tab's views (Insights: 6–8, History: 6–0)
    Tab ↑↓ EnterMove and press; charts take ↑↓ Enter too
    EscBack to the prompt; ctrl+x then Tab returns
    r h s t i p cRun, headless, stop, triage, insights, Playwright code, CI (Run tab)
    v · fEvidence viewer (Insights) · Refresh (Tests, Assurance)

    Settings

    kane-cli commandkane-cli, or any launcher such as node …/dist/index.js
    Default run timeout600 s for pane and Claude runs
    Mirror other kane-cli runsOn: follow terminal, CI and other agents' runs live
    Open the pane at session startnever (default), kane-projects, always

    In Claude Code: /config → kane-qe.

    Getting it

    Install and first run

    Load the mod

    Point Claude Code at the folder in ~/.claude/settings.json → env.CLAUDE_CODE_PLUGIN_DIRS, or start with claude --plugin-dir ~/Documents/kane-claude-plugin/kane-qe.

    Sign in to KaneAI

    npm install -g @testmuai/kane-cli, then ! kane-cli login --oauth from the prompt. CI uses --username / --access-key; Copy CI command writes it for you.

    Grid and devices

    Grid suites need the remote-execution plugin and a HyperExecute MultiOS plan. Local mobile needs macOS on Apple Silicon with Xcode or Android Studio; Setup's doctor lists what's missing.

    Credits

    A run usually costs 10–50 credits. The pane warns under 15, and a run that runs out reads Out of KaneAI credits, never a test failure.