refactor: alias ClawTeam CLI usage to 'oh' and update config for multi-agent coordination

This commit is contained in:
2026-04-03 11:24:09 +07:00
parent e9fd48950c
commit e9a3a623a5
12 changed files with 563 additions and 124 deletions
+166 -22
View File
@@ -27,32 +27,176 @@ Things like:
### TTS
- Preferred voice: "Nova" (warm, slightly British)
- Default speaker: Kitchen HomePod
- Preferred voice. Preferred: "Nova" (warm, slightly British)
```
## Why Separate?
## ClawTeam Multi-Agent Orchestration
Skills are shared. Your setup is yours. Keeping them apart means you can update skills without losing your notes, and share skills without leaking your infrastructure.
> **Important:** ClawTeam CLI command is `oh`. On our OpenClaw setup, ALWAYS explicitly specify `tmux openclaw` as backend+agent — `oh spawn` defaults to `claude` which we don't have.
> Official repo: https://github.com/HKUDS/ClawTeam
## ClawTeam Orchestration
### ⚠️ CRITICAL: Spawn Syntax
- CLI: `clawteam`
- Purpose: Multi-agent team coordination with git worktree isolation
- Default backend: tmux
- Supported agents: claude, codex, openclaw, gemini, kimi, nanobot
- Key commands:
- `clawteam team spawn-team <team> -n <leader>` — create team
- `clawteam spawn --team <team> --agent-name <name> --task "..."` — spawn worker
- `clawteam task create <team> "..." -o <owner>` — create task
- `clawteam task update <team> <id> --status completed` — update status
- `clawteam inbox send <team> <to> "..."` — message agent
- `clawteam board show/attach <team>` — monitor progress
- `clawteam workspace merge <team> <agent>` — merge worktree back
- Profiles: Use `clawteam preset generate-profile` and `--profile` for non-default providers
- Data dir: `~/.clawteam/` (default)
- Requires: Python 3.10+, tmux, git, network/filesystem access for worktrees
```
oh spawn tmux openclaw --team <team> --agent-name <name> --task "<task>"
```
---
**NOT** `oh spawn --team ...` (defaults to `claude` → will fail)
**NOT** `oh spawn tmux claude` (we don't have claude CLI)
Add whatever helps you do your job. This is your cheat sheet.
### Team Setup
```bash
# Create team (you become the leader — set env vars first)
export CLAWTEAM_AGENT_ID="leader-001"
export CLAWTEAM_AGENT_NAME="leader"
export CLAWTEAM_AGENT_TYPE="leader"
oh team spawn-team <team-name> -d "<description>" -n leader
# View team status
oh status
oh board show <team-name>
oh board attach <team-name> # tiled tmux view
```
### Task Management
```bash
# Create tasks (with owner, dependencies, priority)
oh task create <team> "<task description>" -o <owner>
oh task create <team> "<task>" -o <owner> --blocked-by <task-id>
oh task create <team> "Hotfix" -o <owner> --priority high
# Update status
oh task update <team> <task-id> --status pending
oh task update <team> <task-id> --status in_progress
oh task update <team> <task-id> --status completed
oh task update <team> <task-id> --status blocked
# List/filter tasks
oh task list <team> --status pending
oh task list <team> --owner <agent-name>
oh task list <team> --priority high
# Wait for all sub-agents to finish
oh task wait <team>
oh task wait <team> --timeout 300 --poll-interval 10
```
### Spawn Workers (OpenClaw Specific)
```bash
# ✅ CORRECT — always use tmux openclaw
oh spawn tmux openclaw --team <team> --agent-name <name> --task "<task>"
# ✅ With non-default provider/model via profile
oh spawn tmux --profile <profile-name> --team <team> --agent-name <name> --task "<task>"
# ❌ WRONG — defaults to claude
oh spawn --team <team> --agent-name <name> --task "<task>"
oh spawn tmux claude --team <team> --agent-name <name> --task "<task>"
```
### Communication (Inbox)
```bash
# Send message to agent
oh inbox send <team> <to-agent> "<message>"
# Broadcast to all
oh inbox broadcast <team> "<message>"
# Receive (destructive — consumes message)
oh inbox receive <team> --agent <name>
# Peek (non-destructive)
oh inbox peek <team> --agent <name>
# Watch for new messages
oh inbox watch <team> --agent <name>
```
### Profiles & Presets
```bash
# List built-in provider templates
oh preset list
oh preset show <provider-name>
# Generate reusable profile from preset
oh preset generate-profile <preset> <base-agent> --name <profile-name>
# Or use interactive TUI
oh profile wizard
# Test a profile before using in team
oh profile test <profile-name>
# Fix broken profile on fresh machine
oh profile doctor claude
```
### Git Context & Conflict Checks
```bash
oh context log <team>
oh context diff <team>
oh context files <team>
oh context conflicts <team>
oh context inject <team> --agent <name>
```
### Workspace Management
```bash
oh workspace list <team>
oh workspace merge <team> <agent-name>
```
### Snapshots & Recovery
```bash
oh team snapshot <team> --tag <tag-name>
oh team snapshots <team>
oh team restore <team> --snapshot <tag-name>
```
### Worker Loop Protocol
Workers should NOT exit after completing their initial task. Expected loop:
```bash
# 1. Check tasks
oh task list <team> --owner <me>
# 2. Check inbox for new instructions
oh inbox receive <team> --agent <me>
# 3. If idle, notify leader
oh lifecycle idle <team>
```
### Cleanup
```bash
oh team cleanup <team> # kill workers, cleanup worktrees
oh lifecycle request-shutdown <team>
```
### JSON Output
```bash
oh --json team discover
oh --json board show <team>
oh --json task list <team> --status pending
```
Requires:
- `pip install clawteam`
- Python 3.10+
- `tmux` installed
- OpenClaw CLI available on PATH (`openclaw`)
- Git repo for worktree isolation
- Data directory: `~/.clawteam/`