Four model families in Cursor. No fused verdict.
Contents
What this is: a free Cursor skill. The same open question goes to Claude, GPT, Grok, and Gemini in parallel. Limit: it does not pick a winner, it does not check facts, and it does not call Consilium Router.
When it fits
Three short cases. In each one the skill shows the width and does not pick the answer.
A name
Situation: three or four product names are on the table, and one model has already "chosen" its favorite.
What you see: which name several families repeat, and which name only one of them says. A repeated name is not the right name. It is only a sign that the habits lined up.
Split the service
Situation: the question comes before anyone cuts the repository: keep one service, or separate two.
What you see: where the families meet on the boundary, and where one of them sees a coupling the others never name. The report is too early to cut on. It only keeps the disagreement in view.
Critique a plan
Situation: the plan is already written. You want the holes, not a new plan.
What you see: a hole several families name, and a hole only one names. The first is a shared habit. The second may be the only useful one, so the report keeps it, with no score for whose is better.
Lookup, arithmetic, and live prices do not belong here. This run measures nothing and confirms nothing. One fused answer after a panel like this is a Consilium Router deliberation.
What the skill will not do
- It does not pick a winner, and it does not add a recommendation that does.
- It does not silently swap in another model when a seat fails. The roster says who answered, who failed, and who returned nothing.
- It does not pretend to be a deliberation, a vote, a council, a route plan, or a cited pass over a pile of files.
You need Cursor with Task and access to these models. The skill does not need a Consilium key.
How to install
One file, MIT license:
git clone https://github.com/SynenergyAI/consilium-cursor.git
mkdir -p ~/.cursor/skills/consilium-cursor
cp consilium-cursor/consilium-cursor/SKILL.md ~/.cursor/skills/consilium-cursor/SKILL.md
In a new Agent chat: "consilium-cursor: should we split this service?"
Repository: github.com/SynenergyAI/consilium-cursor.
Where a judged answer lives
One judged answer, independent votes, a council, a route plan, or a cited pass over a large set of files is Consilium Router. Each skill run names the Router tool for the next step. It does not call that tool, and it does not invent the answer.