Super Router — AI skill for Claude Code

Super Router

Claude, ChatGPT, Grok and Cursor as one AI team on your Mac. Each task goes to the agent best at it, a second agent checks the work, and a stalled call hands off automatically. Research, red-team, versus and vote modes, task splitting, memory and a local dashboard, all on the AI plans you already pay for.


What it does

You type a task in Terminal (`sr route "…"`) or on a local web page (`sr serve`). Super Router sends it to Claude, ChatGPT, Grok Bot or Cursor and keeps every task in one log.

• One skill per task. Jev, a labeling model, reads each routed task and names its one skill: writing, code, planning, web research, what people are saying on X, reading an image or video you attach, making an image, the same step on many items, a quick answer, and more (config/skill_table.json has the full list). When Jev isn't sure, or isn't set up, you pick the skill.
• One agent does it, another checks it. The skill table names, for each skill, the agent that does the work and a second agent that checks it. If the check finds a problem, the doer gets one round to fix it, and the fix is checked again. Quick answers aren't checked.
• Or pick the agent yourself. Start the text with @claude, @chatgpt, @grok or @cursor. That agent does the work whatever the label. Jev still labels the task to pick who checks it, and if the call stops, it isn't handed to anyone else.
• Teams (presets) put several agents on one task. Each team has default agents, and you can swap them for a single task. Pick a team on the page, or use sr relay in Terminal.
  • research-build: One agent researches the web and X, a second builds from what it found, and a third checks the result.
  • collab: Two agents work as a pair: one plans, the other improves the plan, the first does the work, and the second edits and checks it.
  • versus: Two agents each build their own answer from the task a