Description
arena is an agent skill from Cursor's plugins collection that runs several candidate solutions to the same task in parallel, picks a base, and grafts the strongest parts of the other candidates into it. Its phases are: frame the task, fan out to parallel candidates, cross-judge them, pick a base, graft, and verify. In the framing phase every candidate receives the same prompt, so the prompt acts as the contract. The agent states the artefact each candidate should produce and writes a rubric of three to six concrete, gradeable criteria for success. The rubric is kept for the judging stage, because candidates only see the task. The skill selects the models to run from a configured list in the user's rules, with defaults and family matching when a configured model is refused, and it omits the model setting when the entry means the parent model. The approach is useful for tasks where there are several reasonable designs and you want the evidence to decide between them, rather than trusting the first attempt. It costs more than a single run, so it suits decisions that matter. It is listed on skills.sh, where the page shows install counts and GitHub stars for the Cursor plugins repository, and its security audits all pass. Install it with npx skills add https://github.com/cursor/plugins --skill arena.…
Highlights
- Parallel candidate solutions
- Same prompt as the contract
- Rubric of gradeable criteria
- Cross-judging and picking a base
- Grafting the best parts
- Installs from skills.sh
Ratings & Reviews
No ratings yet
Be the first to rate arena.
People also visit
arena FAQ
What does arena do?
It runs several candidates on the same task in parallel, judges them against a rubric, picks a base and grafts in the strongest parts.
What is the rubric for?
The picker grades candidates with 3 to 6 concrete criteria written before the candidates run.
Which models run?
The skill reads a configured arena runners line in the user's rules, and falls back to defaults if the line is missing.
When is it worth it?
When there are several plausible designs and the evidence should decide, since it runs more than one attempt.
How do I install it?
Run npx skills add https://github.com/cursor/plugins --skill arena.
What are the security audit results?
skills.sh lists three audits: Gen Agent Trust Hub Pass, Socket Pass and Snyk Pass.
What are the best arena alternatives?
Popular alternatives to arena include agent-eval, google-agents-cli-eval and eval-harness-first. See all arena alternatives