Spawns parallel candidate solutions, judges them against a rubric, and synthesizes the strongest result.
npx skills add https://github.com/adjohn/pstack --skill arena-adjohn
Please help me install this Agent Skill. Skill: arena Source: https://github.com/adjohn/pstack/tree/main/skills/arena Command: npx skills add https://github.com/adjohn/pstack --skill arena-adjohn
High-intent search queries and answers about installing and using this skill.
Use the arena workflow: frame a shared prompt with a concrete rubric, spawn N subagents in one message with run_in_background set to true, and assign each its own output path such as a git worktree or /tmp/arena-<slug>/candidate-<n>/.
Spawn one candidate per tier (fable, sonnet, haiku, opus) with the same prompt, then run a readonly cross-judge subagent on a different tier that scores each candidate against the rubric and recommends a base with rationale.
Use parallel candidates when one attempt at a non-trivial artifact would lock in the wrong shape, such as architecture or interface design. Skip the arena for trivial or well-specified tasks where the overhead of N attempts adds no signal.
The arena proceeds with N-1 candidates and records the dropout in the synthesis note. The remaining candidates are still read end to end, scored against the rubric, and used for picking and grafting.
Wild divergence means the framing phase was under-specified: the shared prompt lacked a clear artifact definition or concrete success criteria. The fix is to reframe the prompt and re-run the arena rather than averaging the divergent outputs.
cursorSpawns parallel candidate solutions, judges them against a rubric, and synthesizes the strongest result.
ksciusCoordinate parallel candidate runs and graft best ideas into a verified artifact.
duongbui234Run verification commands and confirm output success before marking development tasks complete.
CasaPerksExecute implementation plan tasks via isolated subagents with spec-first review.
