Claudit

Claude-Led Automated UI-Driven Inspection & Testing

Claudit is a Chrome extension that lets the Claude AI agent directly operate your web admin panels and service pages to run end-to-end (E2E) tests. Drive automated test runs with natural-language instructions, or replay deterministic regression suites written as simple YAML files. Everything happens inside your browser — there is no remote orchestration server. Who this is for QA engineers, test automation engineers, front-end and back-end developers, and product teams who want to replace repetitive manual regression testing with an AI-driven workflow. If you already write test scripts but wish you could describe what needs testing in plain language, Claudit fills that gap. Core features • Natural-language automation — Tell the agent "log in, navigate to the product list, filter by last week, verify the total count" and it performs the clicks, fills, captures, and verifications end to end. • 107 built-in browser tools — click, fill, wait, capture, DOM inspection, network monitoring, and UI-library-aware helpers for AgGrid, Material-UI, Ant Design, PrimeReact, TanStack Table, Handsontable, and more. • Mode 1 — deterministic regression testing: run .test.yaml files against specific URL patterns, with PASS / FAIL results rolled up into an 8-section report. • Mode 2 — team verification: a Planner agent drafts the test plan, an Executor runs each step, and a Verifier independently confirms the result — three-stage validation for high-stakes scenarios. • Spec-grounded testing — Point Claudit at your specification documents (Markdown, YAML, or plain text) and it grounds its verifications against the spec, surfacing discrepancies QA humans often miss. • Local file system integration — Specs, test results, and learning history stay on your local disk via the File System Access API. No cloud sync, no upload. • Per-tab sessions — Each tab is an isolated agent session. Run different tests in parallel tabs without cross-contamination. • Custom project prompts — Register project-specific glossaries, UI conventions, or CRUD patterns that the agent loads automatically. How it works (privacy model) • Everything runs inside your browser. No remote server operated by Claudit, no telemetry, no analytics, no tracking, no cookies. • API calls go directly from your browser to the Claude API endpoint you configured (default: api.anthropic.com). If you point Claudit at an internal proxy or gateway, that is the only outbound destination. • Your Anthropic API key is stored locally via chrome.storage.local. Claudit never transmits it anywhere other than the endpoint you specified. • Sensitive form fields are automatically masked before being sent to the Claude API: input type="password" / "file" / "hidden" values, plus any element with data-sensitive, data-pii, or data-private attributes. • A consent gate (shown on first use) explicitly enumerates what data is transmitted — screenshot, DOM structure, form field values (minus masked fields), page URLs, and optional network request URLs — and blocks the agent until you acknowledge. API usage costs Claudit uses your own Anthropic API key. Every test run calls the Claude API directly from your browser, and the usage fees are billed to your account by Anthropic. Claudit does not collect, relay, or charge any fees. Please review Anthropic's pricing policy before running long test suites. Requirements • A valid Anthropic Claude API key (sign up at console.anthropic.com). • Chrome 120 or newer. • Target sites you own, operate, or have explicit written authorization to test. Scope of use (important) Claudit is intended for testing web applications you own or have been explicitly authorized to test in writing. Do NOT use it for: • Scraping third-party sites you do not own. • Bypassing authentication, rate limits, or anti-automation measures. • Automating accounts that are not yours. • Any activity that would violate a target site's terms of service. Unauthorized use may result in civil or criminal liability (e.g. under the U.S. Computer Fraud and Abuse Act or equivalent local laws). You are solely responsible for ensuring your use of Claudit is authorized. Permissions rationale • activeTab + scripting — read DOM state and dispatch test actions on the tab you explicitly enable. • tabs — read tab metadata (URL, title) for the per-tab session UI shown in the sidebar. • debugger — dispatch synthetic keyboard/mouse events via the Chrome DevTools Protocol. Standard capability used by React DevTools, Lighthouse, Playwright, and similar tools. • webRequest + webNavigation — observe request metadata (URL, method, status) during tests. Request bodies are not read. • storage — save your API key, settings, and test history locally. • sidePanel + windows — provide the chat UI as the extension's side panel. • offscreen — host the File System Access API handle for local spec and memory reads. • notifications — notify you when a long test completes. • alarms — keep the Service Worker awake during long test runs. • host_permissions limited to https://api.anthropic.com/* (for the Claude API fetch). • optional_host_permissions (https://*/*, http://*/*) — you grant per-site access via Options for the sites you actually want to test. Claudit cannot touch any site you have not explicitly added. • optional_permissions (declarativeNetRequestWithHostAccess) — requested only if you configure a custom API endpoint that requires header rewriting. Full documentation • Homepage: https://github.com/youzooyou/claudit • Privacy Policy: https://github.com/youzooyou/claudit/blob/main/PRIVACY_POLICY.md • Terms of Service: https://github.com/youzooyou/claudit/blob/main/TERMS_OF_SERVICE.md • Permissions rationale: https://github.com/youzooyou/claudit/blob/main/docs/permissions.md Claudit is not affiliated with Anthropic. "Claude" is a trademark of Anthropic, used here to describe the API Claudit consumes.