返回 Skill 列表
extension
分类: 开发与工程无需 API Key

spawning-terminals

通过TabzChrome REST API生成和管理终端标签页。在生成工作进程、以编程方式创建终端、为并行工作设置工作树、监控工作进程状态或向Claude会话发送提示时使用。

person作者: jakexiaohubgithub

TabzChrome Terminal Management

Spawn terminals, manage workers, and orchestrate parallel Claude sessions via REST API or MCP tools.

MCP Tools (Preferred)

When TabzChrome MCP server is connected, use these tools instead of raw curl commands:

| Tool | Purpose | |------|---------| | tabz_list_profiles | List terminal profiles (filter by category) | | tabz_list_categories | List profile categories | | tabz_spawn_profile | Spawn terminal using a saved profile | | tabz_get_profile | Get profile details | | tabz_create_profile | Create a new terminal profile | | tabz_update_profile | Update an existing profile | | tabz_delete_profile | Delete a profile | | tabz_list_plugins | List Claude Code plugins | | tabz_list_skills | List available skills |

Profile-Based Spawning with MCP

# List available AI assistant profiles
tabz_list_profiles(category="AI Assistants")

# Spawn using a profile with workingDir override
tabz_spawn_profile(
    profileId="claude-worker",
    workingDir="~/projects/myapp/.worktrees/V4V-ct9",
    name="V4V-ct9"
)

# Check what skills workers will have
tabz_list_skills()

Prerequisites (for REST API)

# Check TabzChrome is running
curl -sf http://localhost:8129/api/health >/dev/null && echo "OK" || echo "TabzChrome not running"

# Get auth token
TOKEN=$(cat /tmp/tabz-auth-token)

Spawn API

Basic Spawn

TOKEN=$(cat /tmp/tabz-auth-token)
curl -X POST http://localhost:8129/api/spawn \
  -H "Content-Type: application/json" \
  -H "X-Auth-Token: $TOKEN" \
  -d '{"name": "Worker", "workingDir": "~/projects"}'

Spawn with Command

curl -X POST http://localhost:8129/api/spawn \
  -H "Content-Type: application/json" \
  -H "X-Auth-Token: $TOKEN" \
  -d '{
    "name": "Build Worker",
    "workingDir": "~/projects/myapp",
    "command": "npm run build"
  }'

Spawn Claude Worker

PLUGIN_DIRS="--plugin-dir $HOME/.claude/plugins/marketplaces --plugin-dir $HOME/plugins/my-plugins"

curl -X POST http://localhost:8129/api/spawn \
  -H "Content-Type: application/json" \
  -H "X-Auth-Token: $TOKEN" \
  -d "{
    \"name\": \"ISSUE-ID\",
    \"workingDir\": \"/path/to/worktree\",
    \"command\": \"BEADS_NO_DAEMON=1 claude $PLUGIN_DIRS\"
  }"

Profile-Based Spawning (Recommended)

Profiles encapsulate command, theme, and settings. Use profiles for consistent worker spawning:

# List profiles
curl -s http://localhost:8129/api/profiles | jq '.profiles[] | {id, name, category}'

# Spawn using a profile via /api/agents
curl -X POST http://localhost:8129/api/agents \
  -H "Content-Type: application/json" \
  -H "X-Auth-Token: $TOKEN" \
  -d '{
    "profileId": "claude-worker",
    "name": "V4V-ct9",
    "workingDir": "~/projects/.worktrees/V4V-ct9"
  }'

Profile-based spawning inherits:

  • Command (e.g., claude)
  • Theme/colors
  • Font settings
  • Environment variables

Spawn Parameters

| Param | Required | Default | Description | |-------|----------|---------|-------------| | name | No | "Claude Terminal" | Display name (use issue ID for workers) | | workingDir | No | $HOME | Starting directory | | command | No | - | Command to auto-execute after spawn | | profileId | No | default | Terminal profile for appearance and command |

Response

{
  "success": true,
  "terminalId": "ctt-ISSUE-ID-abc123",
  "tmuxSession": "ctt-ISSUE-ID-abc123"
}

Worker Management

List All Workers

curl -s http://localhost:8129/api/agents | jq '.data[]'

Find Worker by Name (Issue ID)

curl -s http://localhost:8129/api/agents | jq -r '.data[] | select(.name == "V4V-ct9")'

Get Session ID

SESSION=$(curl -s http://localhost:8129/api/agents | jq -r '.data[] | select(.name == "V4V-ct9") | .id')

Kill Worker

curl -s -X DELETE "http://localhost:8129/api/agents/$SESSION" \
  -H "X-Auth-Token: $TOKEN"

Capture Terminal Output

curl -s "http://localhost:8129/api/tmux/sessions/$SESSION/capture" | jq -r '.data.content' | tail -50

Detect Stale Workers

CUTOFF=$(date -d '5 minutes ago' -Iseconds)
curl -s http://localhost:8129/api/agents | jq -r \
  --arg cutoff "$CUTOFF" '.data[] | select(.lastActivity < $cutoff) | .name'

Sending Prompts via tmux

SESSION="ctt-V4V-ct9-abc123"

# Wait for Claude to initialize (8+ seconds after spawn)
sleep 8

# Find safe-send-keys.sh for reliable prompt delivery
SAFE_SEND_KEYS=$(find ~/plugins ~/.claude/plugins -name "safe-send-keys.sh" -path "*conductor*" 2>/dev/null | head -1)

# Send prompt
PROMPT="Complete beads issue V4V-ct9. Run: bd show V4V-ct9 --json"
"$SAFE_SEND_KEYS" "$SESSION" "$PROMPT"

# Verify delivery
tmux capture-pane -t "$SESSION" -p | tail -5

Git Worktrees for Parallel Workers

Worktrees allow multiple workers on the same repo without conflicts.

Create Worktree

ISSUE_ID="V4V-ct9"

# Create worktree with new branch
git worktree add ".worktrees/$ISSUE_ID" -b "feature/$ISSUE_ID"

Initialize Dependencies

Worktrees share git but NOT node_modules. Initialize before spawning:

INIT_SCRIPT=$(find ~/plugins -name "init-worktree.sh" -path "*conductor*" 2>/dev/null | head -1)
$INIT_SCRIPT ".worktrees/$ISSUE_ID" 2>&1 | tail -5

Full Spawn Workflow

ISSUE_ID="V4V-ct9"
WORKDIR=$(pwd)
TOKEN=$(cat /tmp/tabz-auth-token)
PLUGIN_DIRS="--plugin-dir $HOME/.claude/plugins/marketplaces --plugin-dir $HOME/plugins/my-plugins"

# 1. Create worktree
git worktree add ".worktrees/$ISSUE_ID" -b "feature/$ISSUE_ID"

# 2. Initialize dependencies SYNCHRONOUSLY
INIT_SCRIPT=$(find ~/plugins -name "init-worktree.sh" -path "*conductor*" 2>/dev/null | head -1)
$INIT_SCRIPT ".worktrees/$ISSUE_ID" 2>&1 | tail -5

# 3. Spawn terminal
curl -s -X POST http://localhost:8129/api/spawn \
  -H "Content-Type: application/json" \
  -H "X-Auth-Token: $TOKEN" \
  -d "{
    \"name\": \"$ISSUE_ID\",
    \"workingDir\": \"$WORKDIR/.worktrees/$ISSUE_ID\",
    \"command\": \"BEADS_NO_DAEMON=1 claude $PLUGIN_DIRS\"
  }"

echo "Waiting for Claude to initialize..."
sleep 8

# 4. Get session ID
SESSION=$(curl -s http://localhost:8129/api/agents | jq -r --arg id "$ISSUE_ID" '.data[] | select(.name == $id) | .id')

# 5. Find safe-send-keys.sh
SAFE_SEND_KEYS=$(find ~/plugins ~/.claude/plugins -name "safe-send-keys.sh" -path "*conductor*" 2>/dev/null | head -1)

# 6. Send prompt
PROMPT="Complete beads issue $ISSUE_ID. Run: bd show $ISSUE_ID --json"
"$SAFE_SEND_KEYS" "$SESSION" "$PROMPT"

Cleanup Worktree

ISSUE_ID="V4V-ct9"

# Get and kill session
SESSION=$(curl -s http://localhost:8129/api/agents | jq -r ".data[] | select(.name == \"$ISSUE_ID\") | .id")
curl -s -X DELETE "http://localhost:8129/api/agents/$SESSION" -H "X-Auth-Token: $TOKEN"

# Remove worktree and branch
git worktree remove ".worktrees/$ISSUE_ID" --force
git branch -d "feature/$ISSUE_ID"

Naming Convention

Use the issue ID as the terminal name. This enables:

  • Easy lookup via /api/agents
  • Clear display in tmuxplexer dashboard
  • Correlation: terminal name = issue ID = branch name = worktree name

Worker Dashboard

Launch tmuxplexer watcher to monitor all Claude sessions:

curl -s -X POST http://localhost:8129/api/spawn \
  -H "Content-Type: application/json" \
  -H "X-Auth-Token: $TOKEN" \
  -d '{
    "name": "Worker Dashboard",
    "workingDir": "/home/marci/projects/tmuxplexer",
    "command": "./tmuxplexer --watcher"
  }'

Important Notes

  1. Name terminals with issue ID for easy lookup
  2. Initialize deps SYNCHRONOUSLY before spawning (not background)
  3. Use BEADS_NO_DAEMON=1 in worker command (worktrees share DB)
  4. Pass --plugin-dir flags so workers have plugins
  5. Wait 8+ seconds before sending prompt for Claude to initialize
  6. Workers follow PRIME.md - MCP tools and bd sync work in worktrees