What is Shiplight AI?
Shiplight gives your coding agent eyes and hands in a real browser — your agent becomes the primary author and maintainer of your YAML E2E tests, while humans stay in control by reviewing readable tests and hand-tuning complex flows when needed. It works with Claude Code, Cursor, Codex, and any MCP-compatible agent.
Your agent does the work: /shiplight
Shiplight ships as a single /shiplight skill for your coding agent. Reach for the skill first — scaffolding a project, setting up auth, authoring and fixing tests, wiring CI, and reviewing your app are all things the agent does for you. The rest of this documentation exists so you can understand what the agent produces and hand-tune it when a flow needs human judgment; you rarely need to run these steps by hand.
Invoke /shiplight <subcommand> [context]. Run /shiplight alone (or /shiplight help) for the menu, and /shiplight help <subcommand> to read what one does without running it. Natural phrasing works too — /shiplight screenshot the change resolves to verify.
| Group | Subcommand | What the agent does |
|---|---|---|
| Setup | init | Scaffold a Shiplight test project and write specs/context.md |
auth | Set up or repair login and saved storage state | |
update | Refresh installed Shiplight skills | |
| Author | create-yaml-tests | Plan, scaffold, and write deterministic YAML E2E tests by walking your app in a real browser |
create-agent-verification | Author a reusable agent-driven verification script (browser, API, DB, logs) with auditable PASS/FAIL reports | |
cover | Decide what to test and the strategy at the lowest sufficient cost, drive the producers, and report coverage | |
| Maintain | fix | Reproduce failing or drifted tests, diagnose root cause, repair the YAML, and report app bugs |
| Check | verify | Visually confirm UI changes in the browser during local development |
| Review | review | App-quality review across security, privacy, compliance, design, resilience, performance, SEO, and GEO |
| Ship | ci | Wire CI workflows and the failure-triage pipeline |
cloud | Read Shiplight Cloud (Nova) results — runs, failing/flaky tests, artifacts | |
| Help | help | List or explain subcommands — never executes |
Install the skill and MCP server once — see the Quick Start.
Agent Workflow
Shiplight V2 is built around an agent-native testing workflow:
- Use the browser MCP server and Shiplight Skills with your coding agent
- Let the agent inspect your app in a real browser and verify UI behavior
- Let the agent author and maintain YAML E2E tests — end-to-end tests written in YAML format with natural language steps
- Review tests as readable specs, then hand-tune complex UI flows in the visual debugger
- Keep tests in your repo alongside application code
- Debug test failures locally with the Shiplight CLI and visual debugger
- Run the same YAML E2E tests in GitHub Actions or on Shiplight-hosted CI runners
- Manage CI test results in Shiplight for enterprise teams
How It Works
- Install MCP and Skills so your coding agent can launch browsers, inspect pages, and use Shiplight workflows.
- Verify UI changes by asking the agent to exercise the feature in a real browser.
- Author and maintain YAML E2E tests from the same browser session, using natural language steps that are easy to review.
- Review and tune locally when tests need human control, with screenshots, traces, and a visual debugger for complex UI flows.
- Run in CI through GitHub Actions or Shiplight-hosted CI runners, then review managed test results in Shiplight.
Why This Workflow
- Agent-authored — coding agents do the primary test creation and maintenance work
- Human-controlled — YAML keeps tests easy to review, and the debugger lets you hand-tune complex UI cases
- Repo-native — YAML E2E tests are versioned with your code and run from your project
- CI-ready — local tests run the same way in GitHub Actions and Shiplight CI
- Debuggable — failures include browser evidence, traces, and step-level context
Get started: Setup Guide