arenalisted
Install: claude install-skill HungryKelvin123/dstack
# Arena
Before delegating or using optional capabilities, read [the runtime contract](../david-mode/references/agent-runtime.md). It owns eligibility, model selection, limits, and fallback.
Fan out N isolated attempts at the same bounded task under the runtime worker policy. The parent frames alternatives and invariants, reads every candidate, chooses a base, and integrates the useful parts. Difficult production code and final architecture stay with the parent; workers can supply evidence, bounded sketches, or disposable prototypes for those decisions.
## Start
Open a todolist with one entry per phase before launching anything. The arena runs autonomously and the list keeps phases from silently disappearing.
1. Frame
2. Fan out
3. Cross-judge
4. Pick
5. Graft
6. Verify
## Phase A: Frame
The N candidates will receive the same prompt, so the prompt is the contract. Get it right before spawning anything.
1. State the artifact each candidate is producing and why the requested comparison earns its repeated-context and review cost. Settle shared constraints in the parent before dispatch; candidates must not depend on each other's intermediate work.
2. Derive the rubric. State what success looks like for *this* task, then turn it into 3-6 concrete gradeable criteria. Concrete: `Adds a --dry-run flag that skips writes`. Vague: `code is correct`. The rubric is the picker's tool in Phase D; candidates only see the task.
3. Choose the smallest candidate set that tests distinct p