Sam Corl

You can run, author, and resume browser-based user flows via MCP Playwright, mixing UI steps with direct API calls in one session, and generate self-contained HTML manual QA test plans with interactive pass/fail/pending controls, live progress tracking, and markdown round-trip saving.

They target QA engineers, web developers, and testers who need to verify features in a browser or production environment, manage test cases and plans per project, and hand human testers a clickable checklist with live pass/fail counts and a progress bar.

Both are user-invocable skills configured per project via .claude/user-flows.config.json or .claude/manual-test-plans.config.json. Flows run through MCP Playwright with run state and logs under a configurable directory; test plans are authored in markdown and always regenerated as HTML from that source.

User flows require MCP Playwright for browser automation and a per-project .claude/user-flows.config.json defining the flows directory. Manual test plans require a .claude/manual-test-plans.config.json setting the plan directory; plans are authored as markdown, interactively or from an existing source file.

User flows use targeted checks over full snapshots when a step's breadcrumb is known-good, dispatching Haiku subagents only when a step is new, fails and recovers, or its recorded selector goes stale. Manual plans regenerate HTML from markdown, with Save Results writing annotations back as the new source of truth.