Call Claude Code as a non-interactive coding agent on the same machine as OpenClaw. Use when: (1) You need to delegate coding, refactoring, or bug-fixing tas...
--- name: claude-code-local description: "Call Claude Code as a non-interactive coding agent on the same machine as OpenClaw. Use when: (1) You need to delegate coding, refactoring, or bug-fixing tasks to Claude Code, (2) You want non-interactive code exploration and analysis, (3) You need to run batch code reviews, (4) You're setting up automated CI/CD code checks, (5) You want to use Claude Code as a specialized subordinate for project work while keeping the orchestrator agent focused on coordination." --- # Claude Code Local 调用本机 Claude Code 作为非交互 coding agent。Orchestrator 负责指挥和验收,Claude Code 负责写代码。 ## Quick Reference | 场景 | 命令模板 | |------|---------| | 短任务 | `cd <project> && claude --bare -p "<任务>" --max-turns 10` | | 长任务 | `cd <project> && claude --bare -p "<任务>" --output-format stream-json --verbose --include-partial-messages --max-turns 30` | | 全权限 | `cd <project> && claude --bare -p "<任务>" --dangerously-skip-permissions --output-format stream-json --verbose --include-partial-messages --max-turns 30` | | 只读分析 | `cd <project> && claude --bare -p "<任务>" --permission-mode plan --allowedTools "Read,Glob,Grep,LSP" --max-turns 10` | | JSON 输出 | `cd <project> && claude --bare -p "<任务>" --output-format json --max-turns 10` | ## Key Flags | Flag | Purpose | |------|---------| | `--bare` | Minimal mode, skip hooks/LSP/plugins, faster startup | | `-p` / `--print` | Non-interactive, output to stdout (**do not use PTY**) | | `--max-turns N` | Limit turns: 10 short, 30 long | | `--output-format stream-json` | Streaming output, prevents timeout | | `--output-format json` | Single JSON with `result` field | | `--verbose` | Detailed logging (with stream-json) | | `--include-partial-messages` | Incremental output (with stream-json) | ## Permission Modes Default to the lowest level that accomplishes the task. | Level | Mode | Flags | Use Case | |-------|------|-------|----------| | 1 Read | plan | `--permission-mode plan --allowedTools "Read,Glob,Grep,LSP"` | Code exploration, doc review | | 2 Analyze | default | `--allowedTools "Read,Bash,Glob,Grep,LSP"` | Run diagnostic commands | | 3 Edit | acceptEdits | `--permission-mode acceptEdits --allowedTools "Read,Edit,Write,Bash,Glob,Grep,LSP"` | Modify source files | | 4 Full | bypass | `--dangerously-skip-permissions` | Trusted projects, batch ops | ## Timing Expectations | Complexity | Duration | max-turns | |-----------|----------|-----------| | Simple (回答问题) | 2-5 min | 5-10 | | Medium (探索+单文件) | 5-15 min | 10-20 | | Complex (多文件+测试) | 15-30 min | 20-40 | > **Never kill Claude Code when exec times out.** Check `ps aux | grep claude` first. ## Post-Task Checklist - [ ] Summary of changes - [ ] Modified files list - [ ] Commands executed - [ ] `git status --short` - [ ] `git diff --stat` - [ ] Test results - [ ] Next steps ## References - `references/streaming.md` — Stream-JSON event types and filtering - `references/troubleshooting.md` — Common failures and fixes
don't have the plugin yet? install it then click "run inline in claude" again.
by @clawhub