Shipped with AI. Now let’s actually look at it.

Things that happen

• PR merged at 2:14. Screenshots nowhere.
• The Slovak version exists. Probably.
• Who approved this? The bot did.
• Three agents, four branches, one staging, zero memory of what changed.
• “Works on my machine” is now “works in my context window.”

How it works

1. Acceptance plan

   An ACCEPTANCE.md next to the task says what each screen must achieve. A scenario.yml turns it into steps, branches and criteria the CLI can run.

2. Walk

   A Playwright script, an agent with a browser, or a person walks the flow in a shared browser. Every step is captured the same way: full-page screenshots on desktop and mobile, rendered HTML, console errors, e-mails from the test mailbox.

3. Check

   Countable criteria (elements, texts, prices, metadata, overflow) are checked in the live page with evidence. Criteria that need judgement get a probability from an AI evaluator that reads a reduced text outline of the page; QAwalk turns it into pass, fail or needs review.

4. Canvas and stamps

   The run becomes a canvas: screens, branches, locales, evidence. Reviewers approve a step, return it with a note, or decide a manual criterion. The agent reads the notes with qawalk feedback and fixes them in the next run.

Three stamps. The human is the point.

What the canvas shows

• Desktop and mobile screenshots of every step, e-mails and third-party screens
• Pass, partial or fail per criterion, per locale and viewport, with the evidence
• Page metadata: title, description, canonical, hreflang, Open Graph, JSON-LD
• Branch, commit, author, PR and environment; the scenario’s history and a pixel diff against the baseline
• Replay: step by step with a choice at every decision, for people who do not read canvases
• Links to a specific step and criterion you can send to colleagues

What teams use it for

Fits the tools you already use

• Claude Code and Codexa skill that knows when and how to walk, capture, evaluate and share
• GitHubqawalk share --pr keeps one comment per scenario on the pull request
• CIthe CLI runs headless; a read-only token is enough to read results
• MCPruns, feedback, verdicts and the inbox over MCP at app.qawalk.com/mcp
• E-mail and Slackdigest of what needs you, immediate mail when your run is returned, one Slack channel for the team

Install in three steps

QAwalk runs where your code and your agent are. The service stores and versions the results, shows the canvas and collects approvals.

1. Add the CLI to the project

   npm install --save-dev github:sinfin/qawalk#v0.1.0

   Copy

   Pre-launch: the repository is private. Ask Sinfin for access.

2. Create the config and install the skill

   npx qawalk init --agent claude --org <your-org>
   npx qawalk doctor

   Copy

   The skill is copied to .claude/skills/qawalk.

3. Create your organization and a token

   Create an organization; you become its admin. The start page offers a tester token in one click, later under Settings → API tokens. Put it into the project .env:

   QAWALK_TOKEN=qwk_…
   TYPESAFE_API_KEY=…   # optional: semantic evaluator

   Copy
4. Run and share

   npx qawalk start --scenario docs/tasks/TASK-123/qa/scenario.yml --env local
   npx qawalk walk        # playwright steps; agent/manual steps: qawalk capture
   npx qawalk evaluate
   npx qawalk share --pr

   Copy

   Or ask your agent: “Walk the acceptance plan of TASK-123 with QAwalk on local and share the result.”

Data & security

Pre-launch: free while we test with the first teams, no guarantees yet. Pricing and terms will follow, announced in advance. Pre-launch terms.

Questions we get

What comes next

A weekly “what changed” canvas assembled from runs, check packs for SEO and accessibility, and a GitHub check that turns green when the run is accepted. We tell admins before anything changes that affects them.