XiaoyaoClaw Cross-Tool Commander
Let any Agent-Skills-capable tool (Claude Code / Codex / OpenCode / Trae / DSH, …) command XiaoyaoClaw / OpenClaw multi-agent — dynamic probe + env fill + forced explicit agent + cross-tool.
Product Introduction
OpenClaw Cross-Tool Commander is an Agent Skills skill installed in external tools’ skill dirs (not OpenClaw) so Claude Code, Codex, OpenCode, Trae, DSH, and other agents can directly command your OpenClaw / XiaoyaoClaw multi-agent. Fixes four external-command potholes: openclaw not in PATH (desktop is bundled), env-var gaps that hide real agents, default --agent fallback that misroutes, hard-coded paths that break on reinstall. Solution: multi-tier dynamic probe (Xiaoyao desktop → PATH → vanilla desktop → local prefix); auto-derived OPENCLAW_STATE_DIR / OPENCLAW_CONFIG_PATH for desktop installs; forced explicit --agent (ask if absent, never default); authoritative identity ↔ id mapping from agents list (ask on miss, never guess); works with old openclaw versions (no mcp/migrate) by reading --help output; no tokens/keys stored in skill, gateway creds via env. Supports running a named agent (openclaw agent), cross-channel messages (message send, Feishu/Telegram, …), and gateway status (health).
Core Features
- Dynamic probe — Windows / macOS / Linux multi-tier (desktop → PATH → vanilla → local prefix)
- Env auto-fill — OPENCLAW_STATE_DIR / OPENCLAW_CONFIG_PATH derived for desktop installs
- Forced explicit agent — --agent required; ask if missing, never default
- Dynamic agent-name map — CN/EN/emoji ↔ id authoritative from agents list
- Cross-tool — any Agent-Skills-capable tool (Claude Code / Codex / OpenCode / Trae / DSH)
- Version-tolerant — old openclaw (v2026.3.x) without mcp/migrate still works via --help
- No secrets stored — gateway creds via env inheritance only