Run a project with two Claude Code sessions that check each other. A builder builds one step of the plan at a time, with tests and evidence; an independent reviewer checks the exact commit in its own copy, re-runs everything, and answers PASS, FIX or STOP against a short list of what blocks. An optional tester makes adversarial passes.
安装
npx skills add kl-autonomous/agent-team-skills --skill build-review-loopnpx skills add kl-autonomous/agent-team-skills --skill build-review-loop -a claude-code -gnpx skills add kl-autonomous/agent-team-skills --skill build-review-loop -a codex -ggemini skills install https://github.com/kl-autonomous/agent-team-skills.git --path skills/build-review-loopgh skill install kl-autonomous/agent-team-skills build-review-loopgit clone --depth 1 https://github.com/kl-autonomous/agent-team-skills /tmp/kl-autonomous__agent-team-skills
mkdir -p ~/.claude/skills ~/.agents/skills
cp -R /tmp/kl-autonomous__agent-team-skills/skills/build-review-loop ~/.claude/skills/build-review-loop # Claude Code
cp -R /tmp/kl-autonomous__agent-team-skills/skills/build-review-loop ~/.agents/skills/build-review-loop # Codex, Cursor, Gemini CLI, Copilot, OpenCode