Claude-to-IM Bridge
Bridge Claude Code / Codex to IM platforms — chat with AI coding agents from Telegram, Discord, Feishu, QQ, or WeChat.
Evidence shows: access control via allowed user/channel lists, tool calls require user approval (inline buttons or text commands), tokens stored with chmod 600 and auto-redacted, threat model and incident response in SECURITY.md. Deductions: dependencies not pinned, external effects (e.g., file writes) not explicitly limited, rollback only mentions token rotation, source attribution based solely on MIT license and author name, publisher identity unverified.
Evidence shows: README and SECURITY.md consistent, troubleshooting section provides common issues, doctor command checks environment. Deductions: dependency availability unverified (e.g., @openai/codex-sdk optional), failure messages lack concrete examples.
Evidence shows: supports multiple IM platforms and two AI agents, provides installation and configuration guides, clear platform limitations (e.g., QQ private chat only). Deductions: capability boundaries not explicit (e.g., file operation scope), trigger precision relies on natural language, environment fit not covering all platform details.
Evidence shows: README well-structured, installation instructions detailed, MIT license, known limitations listed. Deductions: version only 0.1.0, no changelog, maintenance responsibility unclear (no contribution guide or issue tracker).
Evidence shows: output is IM messages, has practical value (remote control of coding agents), cost-benefit reasonable (free open source). Deductions: output usability depends on IM platform, marginal value not quantified.
Evidence shows: README claims consistent with code structure, but no independent verification. Deductions: no test results or third-party verification, facts and inferences not clearly separated.
- Dependencies are not pinned, posing supply chain risks.
- Publisher identity unverified, use with caution.
- External effects (e.g., file writes) not explicitly limited, may impact system security.
- Rollback mechanism incomplete, only token rotation mentioned.
What does this agent do, and when should you use it?
Claude-to-IM is a skill that runs a background daemon to connect your IM bots to Claude Code or Codex sessions. Messages from IM are forwarded to the AI coding agent, and responses (including tool use, permission requests, and streaming previews) are sent back to your chat. It supports five IM platforms: Telegram, Discord, Feishu/Lark, QQ, and WeChat, enabling any combination. An interactive setup wizard (`/claude-to-im setup`) guides you through token collection, permission configuration, and validation. Permission control uses inline buttons (Telegram/Discord) or text `/perm` commands with quick `1/2/3` replies (Feishu/QQ/WeChat). The daemon runs in the background and keeps working after you close the terminal. The skill is integrated into Claude Code and Codex, and can be managed via natural language commands after installation.
Runs a Node.js daemon (src/main.ts) that starts the bridge service. It communicates with the AI agent via Claude Agent SDK (src/llm-provider.ts) or Codex SDK (src/codex-provider.ts) using SSE streaming. It receives and sends messages through each IM platform's bot API. It implements a permission gateway (src/permission-gateway.ts) that maps the SDK's canUseTool callback to inline Allow/Deny buttons or text commands, and waits for user responses with a 5-minute timeout. Configuration and credentials are stored in ~/.claude-to-im/config.env with chmod 600. Data is stored in JSON files under data/, and logs are auto-rotated and redacted. It provides management commands (/claude-to-im start/stop/status/logs/reconfigure/doctor) and diagnostic scripts (scripts/doctor.sh).
- A developer on the go wants to quickly approve tool calls and see Claude Code's responses from their phone via Telegram or Discord, without opening a terminal.
- A team shares a Claude Code working session in a Feishu/Lark group, where members send messages and approve permissions via chat.
- A personal user uses WeChat QR login to interact with Claude Code via WeChat messages, managing a local codebase.
- A Codex user wants to send commands and receive results via QQ private chat in a CLI-only setup.
- A user needs the bridge to keep running in the background so they can continue conversations in IM even after closing the terminal.
What are this agent's strengths and limitations?
- Supports five IM platforms (Telegram, Discord, Feishu, QQ, WeChat) with any combination enabled.
- Interactive setup wizard with step-by-step token guidance and validation reduces configuration difficulty.
- Granular permission control via inline buttons or text commands, minimizing accidental operations.
- Streaming preview (Telegram/Discord) shows real-time AI responses.
- Session persistence across daemon restarts preserves context.
- Strong security: credential file permissions 600, logs redacted, no inbound network listeners.
- Requires Claude Code CLI or Codex CLI as backend, cannot work independently.
- QQ supports private chats only, with no inline buttons or streaming preview.
- WeChat uses QR login, single-account only, and no streaming preview.
- Feishu/Lark requires version review and approval before bot works.
- Node.js version >=20 required, may need environment upgrade.
- Documentation lacks full architecture details; source code reading needed for deep understanding.
How do you install or deploy this agent?
Install prerequisites: Node.js >= 20. For Claude Code, use npx skills add op7418/Claude-to-IM-skill or git clone https://github.com/op7418/Claude-to-IM-skill.git ~/.claude/skills/claude-to-im. For Codex, run git clone https://github.com/op7418/Claude-to-IM-skill.git ~/code/Claude-to-IM-skill and then execute bash ~/code/Claude-to-IM-skill/scripts/install-codex.sh. After installation, run /claude-to-im setup in Claude Code or claude-to-im setup in Codex. The wizard guides you through choosing channels (Telegram/Discord/Feishu/QQ/WeChat) and entering tokens.
How do you use this agent?
Run /claude-to-im start in Claude Code or start bridge in Codex to start the daemon. Send a message to your bot in the IM app. When Claude needs tool permission, click the Allow/Deny buttons in chat (Telegram/Discord) or reply with /perm commands and 1/2/3 quick replies (Feishu/QQ/WeChat). Use /claude-to-im status or bridge status to check status, and /claude-to-im logs or 查看日志 to view logs. Run /claude-to-im doctor for diagnostics. To update, use npx skills add op7418/Claude-to-IM-skill or enter the install directory and run git pull && npm install && npm run build.
How does this agent compare with similar options?
The README mentions that this skill is extracted from the CodePilot desktop app, which provides a full-featured GUI with visual chat interface, session management, file tree preview, and permission controls. Users preferring a graphical interface may consider CodePilot instead.
FAQ
Does the bridge support WeChat group chats?
How are tokens kept secure?
~/.claude-to-im/config.env with 600 permissions, logs are redacted, and the daemon has no inbound network listeners.What happens if permission is not approved within 5 minutes?
Can I connect to multiple IM platforms simultaneously?
How do I diagnose connection issues?
/claude-to-im doctor (Claude Code) or doctor (Codex), which checks Node version, config, token validity, logs, and more.