Who it is forA developer with a repo and a coding-agent CLI who wants agents to make a change without watching a terminal.
The problemCoding agents start work without a clear plan, grade their own output and leave reasoning in a long scrollback.
The outcomeA run starts only after you confirm the plan, and its progress appears in one session with gates you can answer.
What you need first
- Node.js 22 or newer and npm 10 or newer
- A container runtime for jobs: Rancher Desktop, Docker, Podman, Colima or OrbStack
- At least one coding-agent CLI you already use, such as Claude Code or Codex
- A modern browser such as Chrome, Edge, Firefox or Safari on macOS
Steps
- Run
npx wicked-crew serveand open the URL it prints, by default http://127.0.0.1:7701. - Open the Agents tab and sign in each coding agent you want to use.
- Choose New job, then Start with chat, and describe the change you want.
- Read the plan the chat proposes and confirm it; nothing starts before you confirm.
- Open the session at /s/:id and answer gate rows as they appear (approve, steer or reject).
Prompt for your agent
In the wicked-studio repo, use the Start with chat flow to describe a small change to <target repo>, then confirm the proposed plan. Expected output: a session URL under /s/ with at least one gate row. Verify that the session header shows the assurance state before you approve any gate.
Caveat. Studio shows gate outcomes and never grades the work itself, so you still review the diff and evidence before accepting a result.
Use this in your AI agent
Ask your agent
Using the Deduped MCP, get the setup steps for mikeparcewski/wicked-studio and "Start a governed agent job from a chat plan"
Add Deduped to Claude Code once
claude mcp add --transport http deduped https://dedupedai.com/mcp
Every Monday: the 5 hottest repos and exactly how to use them. Free, one email a week, unsubscribe anytime.
Get them freeShare: X LinkedIn Hacker News Email Copy link
Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer