Work with OpenL Studio by simply asking your AI assistant in plain language. This plugin connects Claude Code, the Claude desktop app, Cursor, or Codex to OpenL Studio; once connected, the assistant can look things up in Studio for you and act on your behalf — no need to click through Studio screens or know where things live:
- Explore — "Which projects do I have? Show me the tables in the Rating project."
- Understand — "Explain what the LossRatio table calculates, in business terms."
- Investigate — "Where is the final premium discount actually computed?"
- Verify — "Run the tests in this project and explain the failures."
Each tool sets up differently, so start with the guide for yours. Setup takes five to ten minutes.
| The tool you use | Setup guide |
|---|---|
| Claude Code — terminal, IDE extension, or the Claude desktop app's Code tab | docs/claude-code-setup.md |
| Claude desktop app — the Chat or Cowork tabs | docs/cowork-setup.md |
| Cursor | docs/cursor-setup.md |
| Codex — desktop app or CLI | docs/codex-setup.md |
Not sure which you have? If you type your requests into a terminal window or a panel inside your IDE, it's Claude Code. If you use the Claude app with a sidebar of conversations, it's the desktop app — and the tab you're in (Code, or Chat/Cowork) decides which of the first two guides to follow.
Cursor users: install openl from Cursor's public Marketplace once it is
published, from a team marketplace, or with Customize → Plugins → + Add → From
GitHub Repository when your organization's policy permits repository imports. The
Cursor guide explains each route.
Claude in the browser (claude.ai, including cloud sessions) can't be connected: the OpenL connection runs on your own computer, and a browser session has no way to start it. Use one of the tools above instead.
Already set up by your organization? Skip the install step and start at the step of your guide that creates your access token in OpenL Studio. Unless your Studio is single-user (no sign-in screen), that token is yours to create — it isn't something an administrator can hand out for you.
Upgrading from openl-ai 0.1.x? Version 0.2.0 renames the plugin to openl:
docs/migrate-to-0.2.md.
- One of the tools above.
- Node.js 24 or newer on your own computer (the OpenL connection runs locally). Each setup guide explains how to check.
- The OpenL Studio address — the web address you open in your browser to use
OpenL Studio, for example
https://studio.example.com. - Your usual OpenL Studio account, and your office network or VPN if Studio is internal.
Some examples of everyday analyst tasks — they work the same in Claude, Cursor and Codex:
- "List the OpenL projects I can access."
- "Open the Rating project and show me its tables."
- "Explain the DriverRisk table in business terms — what does it decide and based on what inputs?"
- "Where in this project is the final premium calculated? Walk me through the steps."
- "Run the tests in the Rating project and give me a short summary of what failed and why."
- "Compare the AutoPremium table with the version from the previous revision — what changed?"
- "Why did this policy get a premium of 0? Here's the input JSON."
Your assistant picks the right OpenL operation automatically — there are 50+ of them, covering projects, tables, tests, tracing, and deployment.
Skills are guided workflows the plugin adds to your assistant. Pick one explicitly in your client, or just describe the problem — the assistant can start the matching skill by itself.
| Skill | What it does |
|---|---|
openl:connect |
Sets up (or repairs) the connection to OpenL Studio — it is part of the setup guide for your tool (the table above). |
openl:planning |
The entry point for building, changing, or fixing anything in an OpenL project: works out which projects and tables a request touches, shows one plan for a single approval, then carries it out item by item and verifies the result. |
openl:branching |
Manages isolated branches for any rule or config change — opens the right revision, syncs with base/development, resolves conflicts, and handles the hotfix two-branch rule and post-merge cleanup. |
openl:trace-investigation |
Finds out why a rule returned an unexpected result: traces the run, names the root cause, and proposes the minimal fix. Give it the input payload and what you expected. |
openl:change-log-and-reapply |
Keeps a log of every change a feature applied and, after a merge conflict, a project refresh, or a backport, reapplies only the changes that are missing — stopping on conflicting changes in the target. |
openl:testing |
Adds test rows and runs the full suite after any rule change, reading per-row results before reporting anything as tested. |
openl:versioning |
Adds a new effective-dated or dimension-scoped version of a table — without touching the existing version — with matching test coverage. |
To start one explicitly:
- Claude Code or the Claude desktop app's Chat/Cowork tabs: type
/openl:connect,/openl:planning,/openl:branching,/openl:change-log-and-reapply,/openl:trace-investigation,/openl:testing, or/openl:versioning. - Codex: type
$and select$openl:connect,$openl:planning,$openl:branching,$openl:change-log-and-reapply,$openl:trace-investigation,$openl:testing, or$openl:versioning;/skillsopens the skills picker too. - Cursor: the skills are listed under Customize → Skills (in the Agent
Decides section) and can be picked from the
/menu in chat.
The trace workflow behaves the same across clients: it needs a connected Studio and
uses whichever trace tools the configured OpenL MCP server exposes. The connect
workflow instead adapts its setup steps to Claude Code, Cursor, Codex, or Claude
desktop/Cowork.
Upgrading the old manually installed trace skill? After installing plugin 0.3.0, delete
~/.claude/skills/openl-trace-investigation(macOS/Linux) or%USERPROFILE%\.claude\skills\openl-trace-investigation(Windows), then start a new Claude session (or run/reload-pluginsin Claude Code). Plugin updates cannot remove that user-owned copy, and leaving it in place gives Claude two competing trace workflows; the old one targets trace tools that the plugin's pinned server does not expose.
Start with the "If something doesn't work" section of your setup guide — the fixes differ per tool. For more symptoms and fixes, including what to send your administrator: docs/troubleshooting.md.
Still on an old plugin version after a release? By default, updates from this marketplace are manual. Claude Code users can enable auto-update for a marketplace they control, and an administrator may manage it centrally; desktop Chat/Cowork updates remain manual. In Cursor, update behaviour depends on whether the plugin came from the public Marketplace, a team marketplace, or a direct repository import. See The plugin stays on an old version.
- Setup guides, one per tool (the table above): Claude Code, Claude desktop app / Cowork, Cursor, Codex.
- docs/migrate-to-0.2.md — one-time migration from
the
openl-ai0.1.x plugin identity toopenl0.2.0. - docs/troubleshooting.md — symptom → fix, for everyone.
- docs/admin-setup.md — for OpenL administrators: supported versions, organization-wide setup, authentication, security notes.
- docs/architecture.md — for developers: how the plugin is put together.
- docs/release.md — for maintainers: versioning, releasing, distribution.