---
name: "playwright-bowser"
description: "Customer-safe Segmently browser companion for SHOW walkthroughs, screenshots, and generated E2E action contracts. Also trigger when Segmently Launch Assistant returns executeWith.skill/owningSkill=playwright-bowser, mode=show, mode=e2e, or a browserPlan/driverScript/openCommand contract."
---

# Playwright Bowser

Use this skill when a user asks to show where something is in the Segmently UI,
capture a screenshot, verify visible browser state, or run a browser-backed
action contract produced by Segmently Launch Assistant.

This is the public browser companion shipped in the Segmently support plugin.
It is intentionally smaller than the Segmently internal QA skill with the same
runtime id. It uses `playwright-bowser-core` for browser command syntax and
uses generated launch-guide contracts for Segmently-specific actions.

## Preflight

Before live browser work:

```bash
node --version
npm --version
npx --version
segmently auth status
playwright-cli --help
```

If Node/npm/npx are missing, follow the plugin README host prerequisites first.
If Segmently auth is missing, run:

```bash
segmently auth login
```

If browser tooling is missing, follow `playwright-bowser-core` setup.

## SHOW Flow

1. Resolve the user request with `segmently-launch-guide` when the screen or
   setting is ambiguous.
2. Resolve the exact editor URL and source IDs from the request, authorized focus
   or saved context. Ask only for identity still missing after inspection.
3. Open a persistent browser session with `playwright-cli`.
4. Navigate to the target without changing values.
5. Capture a screenshot when it helps the user locate the control.
6. Return the current URL, screenshot path, and concise next step.

Use the packaged `show-runner.mjs` contract when Launch Assistant provides one.

## Published URL Visual Check

When `segmently-cli-guide` or `segmently-launch-guide` returns a canonical
published funnel URL, open that exact URL. Do not substitute `api.segmently.ai`
or guess a default host.

```bash
playwright-cli -s=published-funnel open https://<canonical-domain><publishedUrl> --headed --persistent
```

Use the visible browser as the customer walkthrough surface, then capture a
screenshot or run a read-only smoke path. For paid funnels, do not enter test
card details or complete checkout unless the customer explicitly approves a test
purchase.

## Browser-Backed DO Flow

Only mutate through an explicit generated action contract:

- the owning launch-guide runtime must name the action;
- the project and screen target must be covered by the user's authorization;
- the contract must include the browser driver and verification read.

Run the packaged `e2e-do-runner.mjs` or the exact browser commands returned by
the contract. A mutation is complete only after the verification read passes.

If auth, tool setup, target ids, or verification are missing, stop and ask for
that missing input. Do not claim completion from a screenshot alone.

## Theme and screen authoring review

Accept `theme-authoring` / `animation-authoring` handoffs with exact destination
and the saved-state receipt. Distinguish project template, onboarding snapshot,
canvas screen, editor preview and published version. Inspect preview iframes and
Shadow DOM where present. A canvas tile does not prove the editor has delivered a
fresh save; check identity and safely reopen/remount only after resolving dirty
screen and parent drafts. Never discard or save unrelated draft work implicitly.

For visual authoring check 320, 375 and 430 px, content geometry, overflow, sticky
actions, loaded images and long text. Test actual answer values, selection change,
back and re-entry. Mascot and custom progress checks require a complete movement
cycle plus cleanup, not one screenshot. Native Play is local replay; it does not
prove a remote animation write. For library work inspect an actual authorized
insertion as well as its example preview. Record exactly which states and widths
were observed, and retain screenshots with target/version context.

Use one working browser session and close only owned sessions when finished.
Do not turn a read-only review into publication or a purchase test.

## Boundaries

- Do not read a Segmently source repository.
- Do not use maintainer-only route books or local debug fixtures.
- Do not invent UI selectors or click paths.
- Do not print credentials, cookies, browser storage secrets, or token-like
  values.
- Do not perform broad destructive browser actions.

## Related Skills

- `playwright-bowser-core` for portable browser command syntax.
- `segmently-launch-guide` for request resolution and SHOW/DO contracts.
- `segmently-cli-guide` for read-back verification through public Segmently CLI.
