bocoboho

Wardendeck — Security skill for Claude Code

Security community

Self-hosted IM ↔ local AI coding agent control plane with approvals, audit and isolation (Feishu/Lark + Claude Code).

How to install Wardendeck

This entry records only its repository, not the path inside it, so there is no exact command to give. Open bocoboho/wardendeck and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Wardendeck does

Self-hosted IM ↔ local AI coding agent control plane with approvals, audit and isolation (Feishu/Lark + Claude Code).

Alternatives in Security

ThinkWatch — Self-hosted AI API and MCP gateway for organizations: SSO and RBAC, per-user identity for MCP tool calls, PII 817 ★
Pocketpaw — Your AI agent in 30 seconds 775 ★
Review Artifact — Creates one polished, self-contained HTML artifact from review findings, audit notes, PR feedback, code review 652 ★

README

WardenDeck

Your computer, one chat message away.

WardenDeck is a **self-hosted control plane** that puts the coding agents on your machine behind Feishu/Lark. Send a task from your phone, watch it run live, approve risky tool calls, and keep a tamper-evident record — with no inbound port and no data leaving your machine.

[![Node.js](https://img.shields.io/badge/node-%3E%3D22.13-339933?logo=node.js&logoColor=white)](https://nodejs.org) [![License: MIT](https://img.shields.io/badge/license-MIT-yellow.svg)](LICENSE) [![Status: prototype](https://img.shields.io/badge/status-prototype-orange.svg)](#current-limitations) [![Platform](https://img.shields.io/badge/platform-macOS%20%7C%20Linux-black)](#requirements)

[Quick start](#quick-start) · [Design doc](DESIGN.md) · [Security](#security-model) · [Issues](https://github.com/bocoboho/wardendeck/issues)

Feishu/Lark messages flow through WardenDeck to Claude Code on your machine, with approvals, audit and isolation

What you get

📱 Work from anywhere. Your phone is the terminal — no SSH, remote desktop, or second app.
🛡️ Approve before it acts. Each tool call can pause for your approval: once, 30 minutes, or deny. Only the person who started the task can approve it.
🧾 Prove what happened. Messages, approvals, agent runs, and attachments go into an append-only log secured by a SHA-256 hash chain.
🔒 Keep the agent in a box. On macOS, Claude Code and every tool it spawns run under sandbox-exec, enforced by the kernel.
🧠 Own the whole stack. Per-bot memory (assistant / project / chat / user scopes), encrypted credentials, local SQLite, and a console that listens on 127.0.0.1 only.