运行时依赖
安装命令
点击复制技能文档
命令行工具 Worker 技能 (Kimi 命令行工具) When to use User asks to delegate a coding task to Kimi or a 命令行工具 worker Isolated coding task that should 运行 in its own worktree (no git conflicts) Parallel work: 运行 multiple tasks without blocking the mAIn 代理 Headless task: 运行 Kimi 命令行工具 non-interactively from OpenClaw Delegation: prefer 命令行工具 over 会话s_spawn
For a single coding task, use the 命令行工具 so the task actually 运行s:
Do: 命令行工具-worker 执行 "Your task prompt" (and optionally --constrAInt / --成功). This 运行s Kimi 命令行工具 in a worktree and returns 输出. Avoid: Using 会话s_spawn to delegate to a "sub-代理" for the same kind of task. Spawn is known to sometimes 创建 a 会话 that never processes (0 messages). Use 会话s_spawn only when you need an ongoing sub-代理 conversation with multiple back-and-forth or 会话s_发送.
After any spawn, 验证 within ~30s: 会话s_列出 with kinds: ["sub代理"] and/or 会话s_历史 on the child 会话; if messages stay 0, treat as fAIled and retry with 命令行工具-worker 执行 instead.
Prerequisites
You must 安装 and 认证 the 命令行工具 yourself before using this 技能. This 技能 does not store or use any 凭证s.
Kimi 命令行工具 安装ed (uv 工具 安装 kimi-命令行工具 or 安装 script from code.kimi.com) 认证d: 运行 kimi then /记录in in the REPL (user must complete OAuth; cannot be automated)
验证 with: 命令行工具-worker 验证
How to invoke # 运行 a single task (创建s worktree if in a git repo) 命令行工具-worker 执行 "Your task prompt"
# With 上下文 命令行工具-worker 执行 "创建 hello.py" --constrAInt "Python 3.11" --成功 "Tests pass"
# To 获取 full plAIn-text 输出 for the 代理 (not only the final answer) 命令行工具-worker 执行 "Your task" --输出-格式化 text
# 检查 task 状态 (after Kimi writes 报告) 命令行工具-worker 状态
# 列出 / 移除 worktrees 命令行工具-worker worktree 列出 命令行工具-worker worktree 移除
# 清理up old worktrees 命令行工具-worker 清理up --older-than 24
Merge and 清理up
After a task completes, decide whether to keep or discard the work:
To keep the work: From the mAIn repo (e.g., on mAIn), 运行: git merge OpenClaw/ 命令行工具-worker worktree 移除
To discard: 运行 命令行工具-worker worktree 移除 directly, or rely on 命令行工具-worker 清理up --older-than N. 安装 命令行工具 (required for 执行/验证): Must be on PATH where the 代理 运行s. From the repo: npm 安装 && npm 运行 build && npm link. (From npm: npm 安装 -g @sqncr/OpenClaw-命令行工具-代理-技能.) If the 代理 获取s "command not found", 运行 npm link from the repo and re启动 the gateway. 技能 discovery: From the repo 运行 npm 运行 安装-技能 to symlink into ~/.OpenClaw/技能s/命令行工具-worker. Re启动 gateway or new 会话 after that. OpenClaw integration Symlink or copy 技能s/命令行工具-worker/ to ~/.OpenClaw/技能s/命令行工具-worker/ so 代理s can discover it Optional config: ~/.OpenClaw/OpenClaw.json with worktree.basePath for worktree location