Runs and resumes browser-based user flow test plans via MCP Playwright with per-project configuration.
npx skills add https://github.com/samcorl/qa-flows --skill user-flows-samcorl
Please help me install this Agent Skill. Skill: user-flows Source: https://github.com/samcorl/qa-flows/tree/main/skills/user-flows Command: npx skills add https://github.com/samcorl/qa-flows --skill user-flows-samcorl
.claude/user-flows.config.json, then run test plans against local or production with run <plan> --env=<env>. - Mixed browser and API steps: Test case steps can combine UI actions with direct API calls (type=api) that share the browser session — log in via the UI, then verify server-side state with a GET request, with full request/response cassettes recorded. - Breadcrumb-based reruns: Each passing step records a breadcrumb (selector, URL, or assertion), so repeat runs use cheap targeted checks instead of full page snapshots, dispatching a background Haiku subagent only when a breadcrumb is new, stale, or recovered from failure. - Pause and resume: Run state is tracked in state.json, so an interrupted run can be resumed mid-test-case with resume or resume from step N. - Use Case: Ask to "test checkout in prod" — the Skill loads your prod env config, walks through the login flow with you, executes each test case step in a real browser, logs pass/fail per step, and records cassettes of every API verification. ## Quick Start Ask the assistant to run a user flow test plan against your local environment, and it will walk you through creating the project config file if one does not exist yet.High-intent search queries and answers about installing and using this skill.
Create a test plan markdown file listing test cases, then say "run <plan_name> --env=local". The Skill loads your .claude/user-flows.config.json, drives a real browser via MCP Playwright through each step, and logs pass/fail results per step.
Add a step with a type=api breadcrumb specifying method, path, expected_status, and optional JSON assertions. The request runs through page.request, so it shares the browser session's cookies — log in via the UI first, then verify server state via API.
Yes. The manual auth strategy navigates to your login path, lets you provide credentials or paste a magic link, and waits for confirmation before continuing. Local environments can use a dev-login-endpoint strategy for instant role-based sessions.
Yes. Run state is persisted in state.json under the run directory. Say "resume" to continue from the first non-passed step, or "resume from step N" to re-run the current test case starting at a specific step.
Each passing step records a breadcrumb with its selector or assertion, so reruns verify with a cheap targeted locator check instead of an expensive full browser_snapshot. A Haiku subagent only re-derives breadcrumbs when a step is new, fails and recovers, or goes stale.
It requires MCP Playwright and a per-project config file, so it is built for Claude Code rather than Claude.ai skills-only environments. It is a slower, higher-judgment suite meant to complement fast CI Playwright specs, not replace them.
iliasjennaneExplore websites to map user flows and generate Playwright-based tests.
claretagrapelike32Maps test coverage to GDD critical paths and maintains a regression suite manifest.
cryptopafiAutomate exploratory QA testing of web applications with structured bug reports.
mykaarmaWrite unit, integration, and end-to-end tests with fixtures and mocks.