Control a running OpenWork app from the shell to inspect UI changes and reproduce chat states. Start fresh sessions, select models, send prompts, wait for completion, capture screenshots, and run JavaScript, with no visible window by default.
Drive the running OpenWork app over CDP from the shell, headless by default (nothing opens on screen). Evaluate JS, take screenshots, open a new session, send a prompt with a chosen model, wait for the run to finish. Use when checking a UI change by hand in a world or pnpm dev, or when reproducing a chat state.
Review the instructions the agent will receive before you install.
For verdicts, write or run an evals/core journey or evals/proofs proof (write-a-spec, run-tests). This skill is for checking changes by hand and reproducing states quickly.
Default to the app in headless Chrome. It runs this checkout (Vite hot reload included) and opens nothing on the developer's screen:
pnpm world up dev-app-web --detach -- --browser # prints webUrl and cdp http://127.0.0.1:<port>
pnpm world down dev-app-web # when you are done
Use a desktop window only when the change needs the desktop shell (titlebar, native menus, deep links, the built-in browser pane, updater, quit and relaunch) or the user asks for it. A desktop window opens and takes focus on this computer:
pnpm dev: CDP on http://127.0.0.1:9823.pnpm world up live-desktop -- --allow-shared-state prints cdp http://127.0.0.1:<port> and uses the real installed profile.Set CDP_URL to the address the world printed for every command.
S=.opencode/skills/drive-desktop-cdp/scripts/cdp.mjs
export CDP_URL=http://127.0.0.1:55322
node $S new-session # always start a fresh session first
node $S send "Claude Opus 5.5" "Say hello" # pick the model by its picker name, then send
node $S wait-idle 180 # until the Run task button is back
node $S shot /tmp/after.png # screenshot of the whole page
node $S eval 'document.querySelector("main")?.innerText.slice(-500)'
new-session first. The app can switch sessions on its own between commands.--allow-shared-state uses your real config and history. Anything you send is real.node $S eval 'location.reload(); 1'.sqlite3 ~/.config/openwork/opencode-v2/state/opencode.db "select type, substr(data,1,300) from session_message where session_id='<id>' order by seq".No reviews yet. Tell others what you thought.
No login needed. Share how it worked for you, ask a question, or note where it did not run.
No comments yet. Be the first to write one.
Manage OpenWork feature releases with feature flags, switches that control access to new behavior. Register features before coding, release them by organization and deployment, prepare a global shutdown option, and remove temporary flags after rollout.
Connect OpenCode browser tools to a locally running OpenWork Electron app to inspect its UI, enter a task, and check the response. This skill supports exploration and debugging through CDP, Chrome's remote control protocol.
Check added lines in a change for information identifying customers, prospects, partners, or outside individuals. Report locations and generic remediation without reproducing identities, and escalate cases where the information is already public.
Connects supported AI clients, including Claude Code and Codex, to OpenWork through MCP, a protocol for accessing external tools. It guides account sign-in and verifies that tools for finding and running organization capabilities are available.
Review an outside contributor’s pull request before running its code. Identify concealed behavior, obfuscated code, risky dependencies, and build hooks, with file locations and specific checks for maintainers to follow up.
Create an OpenCode plugin for OpenWork with custom tools and hooks, which run code at specific events. The skill builds the plugin file and adds its entry to the project configuration.