Claw2Claude

claw2claude — OpenClaw Orchestrates Claude Code

Delegates the user's task to the local claude 命令行工具. You (OpenClaw AI) are the 编排器. Claude Code is the executor.

Security & Transparency

This 技能 has broader-than-average 系统 访问 because it bridges two 运行times (OpenClaw and Claude Code). Everything it does is documented here and in the metadata above.

What this 技能 访问es Resource Why 访问 type ~/.OpenClaw/OpenClaw.json Gateway port + auth 令牌, so notifier.py can POST 结果s back to the user's channel Read-only ~/.OpenClaw/凭证s/ JWT expiry 检查 before Claude 启动s — warns if 令牌 is stale; never modifies Read-only ~/.OpenClaw/代理s/mAIn/会话s/会话s.json Find the user's most recently active channel so 结果s go to the right place Read-only / Claude Code working directory; all file changes are intentional 输出s of the delegated task Read + write <技能_dir>/记录s/ Monthly invocation 记录 (timestamp, mode, project, exit 状态, no prompt content beyond first 200 chars) Write Background processes

Two scripts 运行 detached from the mAIn AI turn:

notifier.py — 启动s when Claude 启动s; polls every 2 s for Claude's completion file; 发送s chunked 结果s to the user's channel via the OpenClaw gateway message 工具; exits immediately after delivery (or after 35 min if Claude never finishes). heartbeat.py — used in background mode only; 监控s the background Claude PID; 发送s a notification when Claude exits. Not used in discuss, 执行, or continue modes.

Neither process modifies OpenClaw config, reads 凭证s a second time, or persists beyond the task lifecycle.

--dangerously-skip-权限s usage

All Claude 会话s are launched with this flag. It is necessary because Claude Code is invoked non-interactively (triggered by a chat message, not a terminal), so it cannot display 权限 prompts. The flag's scope is limited to the project directory — Claude is not given broader 系统 访问 than a normal Claude Code 会话 would have.

If you prefer interactive 权限 prompts, 运行 Claude Code directly in a terminal rather than through this 技能.

Required OpenClaw.json changes

Two config changes must be 应用lied before the 技能 works. These are opt-in — the 技能 cannot make these changes itself:

{ "gateway": { "工具s": { "allow": ["message"] } }, "工具s": { "会话s": { "visibility": "all" } } }

gateway.工具s.allow: ["message"] — lets notifier.py call the gateway message 工具 from outside an AI turn to deliver 结果s 工具s.会话s.visibility: "all" — lets the notifier tar获取 the user's channel 会话 (which differs from the 系统 会话 运行ning Claude)

See README.md for the full 设置up 图形界面de.

Activation A. Activate immediately (any one condition is enough) Condition Examples User explicitly asks for Claude "use Claude for this", "let Claude handle it", "hand it to Claude" User 请求s a stronger / smarter 模型 "use a better 模型", "switch to a more capable AI" Task involves an existing Claude Code project User mentions a path that contAIns .OpenClaw-claude-会话.json User explicitly mentions claude 命令行工具 / Claude Code "运行 it with claude", "open a Claude Code 会话" B. Ask first (when you judge the task has these trAIts)

Before 启动ing, ask the user whether to 创建 a Claude project if any of the following 应用ly:

TrAIt How to judge It's an idea or direction User says "I want to build/de签名/plan/write a…" — no concrete single question yet Multiple parts involved Feature 模块s, document chapters, plan phases — not answerable in one reply Multiple iterations expected Discuss → revise → confirm loop, or clearly not a one-shot Q&A A clear project concept exists 应用, 系统, business plan, product spec, docs, course outline, campAIgn plan, etc. User already has related material Mentions an existing directory, codebase, docs, or prior discussion

Fixed phrasing to use when asking:

This looks like a task that will need multiple rounds of discussion or iteration. Would you like to 设置 up a Claude Code project for this? Benefits:

Full conversation 历史 and 上下文 that persists across 会话s Pick up where you left off at any time All files, plans, and code kept in one working directory

[Yes, 创建 project] / [No, handle it directly]

User says yes: ask for (or suggest) a project path, then launch in discuss or 执行 mode User says no: handle it yourself; do not invoke this 技能

If none of the above conditions 应用ly, handle the 请求 yourself. Do not invoke this 技能.

Four Modes Mode When to use Claude 权限s discuss Requirements are unclear; user wants to explore options or plan architecture --dangerously-skip-权限s (all operations allowed) 执行 Task is clear; user wants code written or changed --dangerously-skip-权限s (fully automatic) continue Resuming the previous 会话 in the same mode Inherits the previous mode's 权限s background User wants Claude to 运行 without