agentleFS
muxy-appmuxy-app/muxy/.agents/skills/e2e-test/SKILL.md

e2e-test

Skill2.4k starsChanged todayApache-2.0View on GitHub

Run Muxy's end-to-end release tests in fresh SwiftVM guests (macOS, Ubuntu Wayland, Ubuntu X11), driven visually by parallel low-cost tester subagents. Use before a release, after a risky change, or when the user runs /e2e-test [macos|wayland|x11|linux].

27 other projects use a version of this.

Trending

More talk lately.about the whole repository

Dec 2025: 0 mentionsJan 2026: 0 mentionsFeb 2026: 0 mentionsMar 2026: 0 mentionsApr 2026: 5 mentionsMay 2026: 4 mentionsJun 2026: 0 mentionsJul 2026: 2 mentionsAug 2026: 0 mentionsSep 2026: 1 mentionOct 2026: 6 mentionsNov 2026: 1 mention
Dec 2025Jun 2026Nov 2026
Mentions a month of the muxy-app/muxy repository that our weekly crawl found. There is no adoption data for this file yet.

Most recently taken up by

Who's talking about muxy-app/muxy

The biggest 5 of 19 mentions linking to the repository that our crawl found. Each line is their own headline.

See all 19

Talked about on GitHub, Hacker News and Bluesky (19 times).

On GitHub · 15

10 more on GitHub

On Bluesky · 1

Used with

Projects with it also use unit-test, design-md and 6 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • kevinaimonsterSkill

    unit-test

    单元测试生成助手。帮用户编写单元测试、集成测试、Mock 设计、测试覆盖率提升。当用户说「帮我写单元测试」「写个测试用例」「怎么测这个函数」「mock怎么写」「测试覆盖率」「写测试」「unit test」「test case」「jest测试」「pytest」「vitest」「testing」时触发。关键词:单元测试、测试用例、测试覆盖率、mock、stub、spy、jest、pytest、vitest、mocha、junit、testing-library、断言、边界测试、异常测试、集成测试、e2e测试、测试驱动、代码覆盖率、coverage、test runner、describe、it、expect、assert、fixture、parametrize、测试金字塔

    Also in 4 projects that use this

  • klauswgSkill

    design-md

    Analyze Stitch projects and synthesize a semantic design system into DESIGN.md files

    Also in 3 projects that use this

  • HashiCorpSkill

    changelog

    Add a `.changelog/<PR_NUMBER>.txt` entry for a GitHub Pull Request, commit, and push.

    Also in 3 projects that use this

  • MicrosoftSkill

    playwright-cli

    Automate browser interactions, test web pages and work with Playwright tests.

    Also in 3 projects that use this

  • BlockSkill

    plan

    Draft a TDD, PRD, or lightweight plan in the conversation without committing it.

    Also in 3 projects that use this

  • GoogleSkill

    improve-codebase-architecture

    Scan a codebase for deepening opportunities, present them as a visual HTML report, then grill through whichever one you pick.

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

8 rules. 0 are common, 4 are distinctive, and 4 are very specific to this project.

  • 0

    Common

    Most of the 23 edited versions we read keep these.

  • 4

    Distinctive

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read first: docs/product/README.md, docs/tech/README.md, and the tester protocol (you hand it to every tester).

    line 20
  2. Distinctive:

    Never let two testers touch the same VM at once.

    line 76
  3. Very specific:

    Start batch A for every platform in one message.

    line 89
  4. Very specific:

    Read the summary block. Any FAIL, BLOCKED or INSTRUCTION row needs your review.

    line 111
  5. Distinctive:

    Ask the user before cleaning up: Ask whether they want any of the run's VMs kept to inspect a failure or reproduce a fix, and wait for the …

    line 146
  6. Distinctive:

    Clean up: When the user is done, delete every VM of the run and confirm none are left:

    line 151
  7. Very specific:

    Not covered: remote servers and pairing (need a second machine), mobile, muxy worktree checkout-pr (needs a GitHub remote), Quick Terminal …

    line 184
  8. Distinctive:

    Write steps a model can follow without judgment: one action per step, exact keys for both platforms, an exact expected result, and a check …

    line 192

“Common” is measured: kept by at least 6 in 10 of the 23 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Muxy E2E tests

You are the coordinator. You never drive a VM yourself. Low-cost tester subagents boot the guests, build the current working tree, and run the suites in suites/ through Agent Desktop (keys, typing, clicks, screenshots) and Muxy's CLI. You plan the run, dispatch testers, review their evidence, and report.

Other skills reuse this harness for one scenario: write the steps as a suite file in the same format, and hand it to testers with the prompt in step 4. The tester protocol carries every rule for driving the VM, including Agent Desktop, so don't write VM rules into a tester's prompt.

Read first: docs/product/README.md, docs/tech/README.md, and the tester protocol (you hand it to every tester).

1. Scope

ArgumentPlatforms
empty, allmacos, wayland, x11
linuxwayland, x11
macos, wayland, x11, windowsthat one

Anything else: stop and show this table.

2. Prepare (host only, two commands)

.agents/skills/e2e-test/scripts/e2e preflight
.agents/skills/e2e-test/scripts/e2e prepare        # prints the run id, e.g. r10302015

preflight fails when a template is missing: follow references/templates.md before testing. It also fails when the host DHCP pool for SwiftVM is full. Each new VM takes a lease, and macOS never gives an expired lease to a new MAC, so the pool runs out after about 250 VMs. A guest then gets no IPv4 address, and its build can't reach github.com. Ask the user to run e2e dhcp-prune in their own terminal, because it needs sudo. up also stops with ENV: when a guest has no network. up also fails with ENV: when Agent Desktop is missing, not ready, or can't take screenshots in the clone; that is a template defect, fixed by refreshing the template. If preflight lists VMs from an earlier run, ask the user whether to clean them up (e2e clean <old run>) before starting; they hold host memory and disk. prepare exports the working tree (tracked edits and new files included) to target/e2e/<run>/source.tar; prepare <run> <checkout> exports another checkout, such as a worktree. If you edit source later, run a new prepare and a new run; never mix sources within a run.

3. Pick the tester model

Choose the cheapest model that has a shell and can read PNG images; the tests are mechanical and need no reasoning beyond comparing what is seen with what is written. Prefer, in order:

HarnessTester
Claude Codemodel: "haiku" on a general-purpose subagent
Codex / ChatGPT agentsgpt-6-luna (or the cheapest listed mini model), low reasoning effort, fork_turns: "none"
Anything elsethe cheapest available model that has shell and image input

Set the model per subagent; never change the user's global settings. If no cheaper model exists, use your own model for testers, but still delegate.

4. Dispatch

Each platform has its own VM and runs its batches in order; the three platforms run in parallel. One tester per batch, fresh context each time. Never let two testers touch the same VM at once.

BatchSuitesNotes
Aboot, S01, S02Boots the VM with e2e up
BS03, S04
CS05, S06
DS07, S08
ES09, S10
FS11
GS12, S14Terminal UI and terminal images
HS13, S15Extensions, AI and agent dialogs, language packs

Start batch A for every platform in one message. When a platform's batch finishes, start that platform's next batch right away; don't wait for the other platforms. If batch A reports BOOT: FAIL, stop that platform. However a run ends, finish with step 6.

Give each tester exactly this prompt, filled in:

You are a Muxy E2E tester. Work only through the commands in the tester protocol.
Read and follow: .agents/skills/e2e-test/references/tester.md
Run id: <run>    Platform: <macos|wayland|x11|windows>    VM: muxy-e2e-<run>-<platform>
Batch: <A-F>     Boot first: <yes for batch A, otherwise no>
Suites, in order: .agents/skills/e2e-test/suites/<file>.md, ...
Return your report as your final message.
Do not spawn agents, edit repository files, or debug Muxy's source.

5. Review

Testers' verdicts are evidence, not conclusions. Save each returned report as target/e2e/<run>/<platform>/report-<batch>.md. For every report:

  1. Read the summary block. Any FAIL, BLOCKED or INSTRUCTION row needs your review; a PASS row with no evidence path is treated as BLOCKED.
  2. For each problem row, open its screenshots and log excerpts yourself and classify it:
    • Product bug: Muxy behaved against the suite's expectation and the docs. Keep it.
    • Environment: VM, template, or tooling failure. Fix the template or script if it is routine (see templates.md), otherwise report it.
    • Test defect: the suite's steps or expectation are wrong. Fix the suite after the run, and say so in the final report.
    • Flaky: re-run only that test once with a new tester (same VM, after its batch) and keep both results.
  3. Spot-check one PASS screenshot per suite to keep testers honest.

6. Finish (mandatory)

A run is not finished until its VMs are gone. Do these steps in order, every time, even when the run failed or was stopped early.

  1. Stop the VMs. Once every batch has reported, stop each platform's VM. This keeps its disk for inspection and frees the host:

    .agents/skills/e2e-test/scripts/e2e down muxy-e2e-<run>-<platform>
  2. Report to the user:
    • Source: commit, dirty file count, archive digest (target/e2e/<run>/source.txt).
    • A platform × suite table: pass / fail / blocked counts.
    • Each confirmed product bug: test id, platform(s), expected vs actual, the key screenshot and log paths, and how reproducible it was.
    • Environment problems and test defects, separately, with what you changed.
    • What is not covered (see the coverage table below).
    • The run's VMs, from e2e vms <run>.
  3. Ask the user before cleaning up. Ask whether they want any of the run's VMs kept to inspect a failure or reproduce a fix, and wait for the answer. Don't delete anything before they reply. If they want some kept, say which VMs remain and ask again once they say they are done.
  4. Clean up. When the user is done, delete every VM of the run and confirm none are left:

    .agents/skills/e2e-test/scripts/e2e clean <run>
    .agents/skills/e2e-test/scripts/e2e vms <run>

    clean copies the logs into target/e2e/<run>/ first, so the evidence stays. Tell the user the VMs are deleted (they are in the Trash). Delete only this run's VMs; never another run's, a template, or a VM the run didn't create.

Coverage

SuiteFeature
S01Build identity, first launch, server auto-start
S02Terminal I/O, Unicode and color, Ctrl+C, scrollback, find, paste, shell integration, copy, font size
S03Splits, focus, zoom, close confirmation, tabs, rename, detach and Existing Terminals
S04Add project, Git status, CLI sessions in the app, rename, switch, delete
S05Terminals outlive the app, restore on launch, server stop and reconnect, End All Sessions and Quit
S06CLI input into a visible terminal, terminal UI, shared sessions, program status, agent activity
S07Project layouts, worktrees from CLI and UI
S08Settings window, saved preferences, theme picker, sidebar, command palette
S09Built-in browser, CLI automation, address bar, back
S10Composer send
S11Worktrees end to end: discovery, form and CLI creation, setup and teardown commands, location templates, register, switch, rename, dirty and clean removal, lost folders, hiding, removing the parent
S12Terminal UI: isolated server, idle output limit, splits, suspend and resume, detach and restore, resize, two windows sharing one layout
S13Extensions: store listing, load, enable, panel, reload, broken package, disable and unload, floating panels
S14Terminal images: Kitty, iTerm and Sixel, file and temp-file transfers, the named-pipe guard, DA1, restore after restart, scrolling
S15Commit dialog, Create Extension with an agent, Linux install of the muxy command, symbol language packs

Not covered: remote servers and pairing (need a second machine), mobile, muxy worktree checkout-pr (needs a GitHub remote), Quick Terminal (needs a recorded global shortcut), dictation, extensions marketplace (network), updates, AI commit/PR (needs signed-in tools), Windows.

Changing the suites

Write steps a model can follow without judgment: one action per step, exact keys for both platforms, an exact expected result, and a check that doesn't rely on typed text (see "Markers" in the tester protocol). Click with $E click-element $VM <name>, using the control's accessible name (its aria_label, or the English text of its tr! label), and use $E click only for what isn't a control, saying why in that step. Never click by accessibility ID (project-row-alpha, settings-toggle-...): swiftvm control reports only each control's role, name and position, not Muxy's IDs, so an ID never matches. When a name repeats, add --role <role> (as look prints it: button, toggle, textfield, item, tab, ...; use the Linux names: toggle also matches macOS's checkbox, and tab also matches macOS's radio) or --nth <n> (the nth match in tree order), and say in the step why. --right finds the control by name and right-clicks it with the pointer. click-element fails and lists the matches when a name is ambiguous. For a native dialog on macOS, ask for $E desktop $VM look --app muxy-app as well as a shot (swiftvm control has no snapshot action). Verify new steps in a guest before committing them. Keep references/tester.md the single place for shared rules.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.