← ClaudeAtlas

agent-comparelisted

Compare multiple external-agent outputs and let Codex produce the final decision.
leviuszen/agent-workbench · ★ 2 · AI & Automation · score 73
Install: claude install-skill leviuszen/agent-workbench
# Agent Compare Use this skill when two or more agent outputs disagree or when Codex needs to decide which result to trust. ## Rules - Compare against the user's original request and local project constraints. - Prefer tests, diffs, and file evidence over confident prose. - Identify partial reuse opportunities instead of picking only one winner. - Write the final decision to `codex-final.md` when working inside a task folder. ## Decision Format Use this structure: - Decision - Why - Accepted parts - Rejected parts - Files or patches to apply - Tests to run - Remaining risks