name: tlive
description: |
IM bridge for AI coding tools — chat with Claude Code / Codex from
Telegram, Discord, or Feishu. Approve permissions, get streaming responses,
manage sessions from your phone.
Use for: starting IM bridge, configuring IM platforms, checking status,
diagnosing issues.
Trigger phrases: "tlive", "IM bridge", "消息桥接", "手机交互", "启动桥接",
"连接飞书", "连接Telegram", "诊断", "查看日志", "配置".
Do NOT use for: building bots, webhook integrations, or general coding tasks.
argument-hint: "setup | stop | status | logs [N] | reconfigure | doctor"
allowed-tools:
Bash
Read
Write
Edit
AskUserQuestion
Grep
Glob
TLive — IM Bridge Skill
You are managing the TLive IM Bridge — bidirectional chat with AI coding tools from Telegram, Discord, or Feishu.
The Bridge uses the Claude Agent SDK (or Codex SDK) to interact with the AI coding tool. It is completely independent from the optional Go Core web terminal server.
Disambiguation: status vs doctor — Use status when the user just wants to check if the bridge is running. Use doctor when the user reports a problem or suspects something is broken. When in doubt and the user describes a symptom (e.g., "没反应了", "挂了"), prefer .
Terminal Live — monitor and chat with AI coding agents (Claude Code, Codex) from Telegram, Discord & Feishu.
Three features, use any combination:
| Feature | What it does | Access from |
|---------|-------------|-------------|
| Web Terminal | tlive <cmd> — wrap any command with a web-accessible terminal | Browser / Phone |
| IM Bridge | /tlive — chat with Claude Code from your phone | Telegram / Discord / Feishu |
| Hook Approval | Approve Claude Code permissions from your phone | Telegram / Discord / Feishu |
Quick Start
# 1. Install
npm install -g tlive
# 2. Configure your IM platform (interactive wizard)
tlive setup
# 3. Register hooks + Claude Code skill
tlive install skills
# 4. In Claude Code, start the bridge
/tlive
Recommended: Run /tlive setup inside Claude Code for an AI-guided setup experience that walks you through each step.
First-time install walkthrough for Claude Code, Codex CLI, and ChatGPT.
doctor
Runtime Detection
AskUserQuestion available → Claude Code → interactive wizard
Not available → Codex / other → show config example, non-interactive
Config Check (all commands except setup)
Before any command except setup, check ~/.tlive/config.env:
Missing → Claude Code: auto-start setup wizard. Codex: show ~/.tlive/docs/config.env.example and stop.
Exists → proceed
Subcommands
/tlive (no args) or start — Start Bridge
1. Check config.env → if missing, auto-start setup
2. Check Bridge PID → if running, show status instead
3. Start Bridge: tlive start (uses TL_RUNTIME from config, default: claude)
4. Wait 2s, verify alive: tlive status
5. Report runtime, channels + web terminal status
setup
Interactive wizard. Collect one field at a time, confirm each (mask secrets to last 4 chars).
Before asking for platform credentials, read ~/.tlive/docs/setup-guides.md internally. Only mention the specific next step the user needs — don't dump the full guide. Show the relevant guide section only if the user asks for help.
Step 1 — Choose IM platforms:
AskUserQuestion: "Which IM platforms to enable?
1. Telegram — streaming preview, inline permission buttons
2. Discord — team use, channel-level access control
3. Feishu (飞书) — streaming cards, tool progress
Enter numbers (e.g., 1,3):"
Step 2 — Collect credentials per platform:
Telegram: Bot Token → confirm (masked) → Chat ID (optional) → Allowed User IDs (optional). Important: At least one of Chat ID or Allowed User IDs should be set.
Discord: Bot Token → confirm (masked) → Allowed User IDs → Allowed Channel IDs (optional). Important: At least one of Allowed User IDs or Allowed Channel IDs should be set.
Feishu: App ID → confirm → App Secret → confirm (masked) → Allowed User IDs (optional).
Step 3 — General settings:
Runtime: claude (default) or codex
Port (default 4590)
Public URL (optional, for web links in IM messages)
Auto-generate TL_TOKEN (32-char hex)
Step 4 — Write config and validate:
Read ~/.tlive/docs/config.env.example as the template — use its exact variable names (e.g., TL_TG_* for Telegram, TL_DC_* for Discord, TL_FS_* for Feishu). Do NOT invent variable names.
Show a summary table (secrets masked to last 4 chars)
Ask user to confirm before writing
mkdir -p ~/.tlive/{data,logs,runtime}
Write ~/.tlive/config.env using the template's variable names, then chmod 600
Validate tokens — read ~/.tlive/docs/token-validation.md for exact commands per platform
Report results. If validation fails, explain what's wrong.
On success: "Setup complete! I'll start the Bridge now." Then auto-start.
reconfigure
Read current config from ~/.tlive/config.env
Show current settings in a table (secrets masked to last 4 chars only)
Ask what the user wants to change
Collect new values one at a time, show where to find each value (show full guide from ~/.tlive/docs/setup-guides.md only if asked)
Update config file
Re-validate any changed tokens
Remind: "Run /tlive stop then /tlive start to apply changes."
stop
tlive stop
status
tlive status
logs
Extract optional line count N from arguments (default 50).
tlive logs [N]
doctor
Run diagnostics and suggest fixes. For complex issues, read ~/.tlive/docs/troubleshooting.md.
tlive doctor
Then validate IM tokens if configured — read ~/.tlive/docs/token-validation.md for commands.
help
Show a clear overview of the TLive system and available commands:
TLive — Three features, use any combination:
In Claude Code (/tlive):
/tlive Start IM Bridge (chat from phone)
/tlive setup Configure IM platforms (AI-guided)
/tlive reconfigure Modify specific config fields
/tlive stop Stop Bridge
/tlive status Show Bridge + Web Terminal + Hooks status
/tlive logs [N] Show last N log lines
/tlive doctor Diagnose issues + suggest fixes
In terminal (tlive):
tlive <cmd> Wrap command with web terminal (e.g. tlive claude)
tlive setup Configure IM platforms (interactive)
tlive install skills Install /tlive skill + hooks to Claude Code
tlive hooks pause Auto-allow permissions, no IM notifications
tlive hooks resume Resume IM approval flow
In IM (from phone):
/new Start new conversation
/model <name> Switch model (e.g. claude-sonnet-4-6)
/runtime claude|codex Switch AI provider
/settings user|full|isolated Claude settings scope
/perm on|off Permission prompts on/off
/effort low|medium|high|max Thinking depth
/stop Interrupt execution
/verbose 0|1 Detail level (quiet/terminal card)
/sessions List recent sessions
/session <n> Switch to session
/hooks pause|resume Toggle hook approval
/status Check status
/help Show commands
Notes
Always mask secrets in output (show only last 4 characters)
Always check for config.env before starting — without it the daemon crashes and leaves a stale PID file
Bridge and Go Core web terminal are independent — Bridge works without Go Core
Go Core is started separately via tlive <cmd> in a terminal, not by this skill
Config at ~/.tlive/config.env — shared by both Bridge and Go Core
The daemon runs as a background Node.js process
tlive claude # Wrap Claude Code
tlive python train.py # Wrap a training script
tlive npm run build # Wrap a build
$ tlive claude --model opus
TLive Web UI:
Local: http://localhost:4590?token=abc123
Network: http://192.168.1.100:4590?token=abc123
Session: claude (ID: a1b2c3)
Multiple sessions in one dashboard. Daemon auto-starts, auto-shuts down after 15 minutes idle.
IM Bridge
Chat with Claude Code from your phone. Start new tasks, get streaming responses with real-time tool visibility.
You (Telegram): "Fix the login bug in auth.ts"
TLive: ● Read(auth.ts)
● Grep("validateToken" in src/)
● Edit(auth.ts)
├ Applied
● Bash(npm test)
├ All 42 tests passed
━━━━━━━━━━━━━━━━━━
Fixed the login bug. The token validation
was missing the expiry check...
📊 12.3k/8.1k tok | $0.08 | 2m 34s
Approve Claude Code tool permissions from your phone. Never get blocked by a [y/N] prompt again.
Claude Code runs normally in your terminal (no wrapper needed)
│
├── Claude wants to run a command
│ → Hook fires → Go Core receives → Bridge sends to your phone:
│
│ 🔒 Permission Required
│ Tool: Bash
│ ┌──────────────────────────┐
│ │ rm -rf node_modules && │
│ │ npm install │
│ └──────────────────────────┘
│ [✅ Yes] [✅ Allow Bash(npm *)] [❌ No]
│
├── You tap [Yes] → Claude Code continues
│
└── Walk away. Claude keeps working.
Phone buzzes only when approval needed.
Safe by design:
If Go Core is unreachable, hooks pass through locally (zero impact on normal usage)
Once Go Core accepts a PermissionRequest, timeout/error defaults to deny (not allow)
Shows exact tool name and command before you approve
Works with any Claude Code session, no wrapper needed
Pause when you're at your desk:
tlive hooks pause # Auto-allow everything
tlive hooks resume # Back to IM approval
Settings scope: By default, bridge only loads ~/.claude/settings.json (API key, model config). Set TL_CLAUDE_SETTINGS=user,project,local or use /settings full in chat to also load project CLAUDE.md, MCP servers, and skills.
Remote Access (frp / tunnels)
To access the web terminal from outside your LAN (e.g. via frpc, Cloudflare Tunnel, ngrok):
Forward local port 4590 (or your TL_PORT) through the tunnel
Set TL_PUBLIC_URL to your tunnel domain:
TL_PUBLIC_URL=https://your-domain.com
IM messages will use this URL for web terminal links instead of the LAN IP.
Security notes: The tunnel exposes full terminal access. Make sure:
TL_TOKEN is set (auto-generated by tlive setup) — all requests require this bearer token
IM user whitelists are configured (TL_TG_ALLOWED_USERS, TL_DC_ALLOWED_USERS, etc.)
Use HTTPS on the tunnel side (frps/Cloudflare handle this automatically)
The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands.
