@songsid/agend 0.0.1

package/README.md

@@ -0,0 +1,210 @@
+<p align="center">
+  <h1 align="center">AgEnD</h1>
+  <p align="center">
+    <strong>Run a fleet of AI coding agents from your phone.</strong>
+  </p>
+  <p align="center">
+    <a href="https://www.npmjs.com/package/@suzuke/agend"><img src="https://img.shields.io/npm/v/@suzuke/agend" alt="npm"></a>
+    <a href="LICENSE"><img src="https://img.shields.io/badge/License-MIT-blue.svg" alt="License: MIT"></a>
+    <a href="https://nodejs.org"><img src="https://img.shields.io/badge/Node.js-%3E%3D%2020-green.svg" alt="Node.js >= 20"></a>
+  </p>
+</p>
+
+AgEnD (**Agent Engineering Daemon**) turns your Telegram or Discord into a command center for AI coding agents. One bot, multiple CLI backends, unlimited projects — each running as an independent session with crash recovery and zero babysitting.
+
+<p align="center">
+  <code>You → Telegram/Discord → AgEnD → Fleet of AI Agents → Results back to your phone</code>
+</p>
+
+[繁體中文](README.zh-TW.md) · [Documentation](docs/features.md) · [CLI Reference](docs/cli.md)
+
+---
+
+## Why AgEnD?
+
+| Without AgEnD | With AgEnD |
+|---|---|
+| Close the terminal, agent goes offline | Runs as a system service — survives reboots |
+| One terminal = one project | One bot, unlimited projects running in parallel |
+| Long-running sessions accumulate stale context | CLI auto-compact + crash recovery with context snapshots |
+| No idea what your agents are doing overnight | Daily cost reports + hang detection alerts |
+| Agents work in silos, can't coordinate | Peer-to-peer collaboration via MCP tools |
+| Runaway costs from unattended sessions | Per-instance daily spending limits with auto-pause |
+
+## Feature Highlights
+
+🚀 **Fleet Management** — One bot, N projects. Each Telegram Forum Topic is an isolated agent session.
+
+🔄 **Multi-Backend** — Claude Code, Gemini CLI, Codex, OpenCode, Kiro CLI. Switch or mix freely.
+
+🤝 **Agent Collaboration** — Agents discover, wake, and message each other via MCP tools. A General Topic routes tasks to the right agent using natural language.
+
+📱 **Mobile Control** — Approve tool use, restart sessions, and manage your fleet from Telegram inline buttons.
+
+🛡️ **Autonomous & Safe** — Cost guards, hang detection, model failover, and crash recovery keep your fleet running without babysitting.
+
+⏰ **Persistent Schedules** — Cron-based tasks backed by SQLite. Survives restarts.
+
+🎤 **Voice Messages** — Talk to your agents with Groq Whisper transcription.
+
+📄 **HTML Chat Export** — Export any agent session as a self-contained HTML file for sharing or archiving.
+
+🪞 **Mirror Topic** — Cross-instance visibility. Watch another agent's work in real time from a separate topic.
+
+🖥️ **Web Dashboard** — Live fleet monitoring in the browser with SSE updates and integrated chat UI.
+
+🔌 **Extensible** — Discord adapter, webhook notifications, health endpoint, external session support via IPC.
+
+👥 **Teams & Task Board** — Named groups for targeted broadcasting. Shared task board for multi-step work tracking across instances.
+
+📋 **Fleet Templates** — Define reusable fleet configurations. Deploy multi-instance setups with one command, each with its own git worktree.
+
+## Quick Start
+
+One-liner (macOS / Linux — installs Node.js via nvm + tmux + agend, then runs quickstart):
+
+```bash
+curl -fsSL https://suzuke.github.io/AgEnD/install.sh | bash
+```
+
+Or install manually:
+
+```bash
+# Option A: One-line install (Linux / macOS / WSL)
+curl -fsSL https://suzuke.github.io/AgEnD/install.sh | bash
+
+# Option B: Manual install
+npm install -g @suzuke/agend    # 1. Install
+agend quickstart                # 2. Setup — bot token, backend, done
+agend fleet start               # 3. Launch your fleet 🎉
+```
+
+Open Telegram, send a message to your bot, and start coding from your phone.
+
+> **Discord?** `agend quickstart` supports Discord too — install the plugin first: `npm install -g @suzuke/agend-plugin-discord`. See [Discord setup guide](docs/features.md#discord-adapter-mvp).
+
+## How It Works
+
+```mermaid
+graph LR
+  You["You<br/>(Phone / PC)"] <-->|messages| Channel["Telegram / Discord<br/>/ Web UI"]
+  Channel <-->|routing| Daemon["AgEnD Daemon"]
+
+  subgraph Fleet
+    Daemon --> General["General<br/>Dispatcher"]
+    Daemon --> A["Instance A<br/>Claude Code<br/>Project X"]
+    Daemon --> B["Instance B<br/>Gemini CLI<br/>Project Y"]
+    A <-.->|MCP Tools| B
+    General -.->|routes tasks| A
+    General -.->|routes tasks| B
+  end
+```
+
+1. **You send a message** to your Telegram/Discord bot
+2. Messages sent to the **General Topic** are interpreted and routed to the right agent. Messages to a specific topic go directly to that instance.
+3. **Agent instances** run in isolated tmux sessions, each with its own project and CLI backend
+4. **Agents collaborate** peer-to-peer via MCP tools — delegating tasks, sharing context, reporting results
+5. **Results flow back** to your chat. Permission requests arrive as inline buttons.
+
+## Supported Backends
+
+| Backend | Install | Auth |
+|---------|---------|------|
+| Claude Code | `curl -fsSL https://claude.ai/install.sh \| bash` | `claude` (OAuth) or `ANTHROPIC_API_KEY` |
+| OpenAI Codex | `npm i -g @openai/codex` | `codex` (ChatGPT login) or `OPENAI_API_KEY` |
+| Gemini CLI | `npm i -g @google/gemini-cli` | `gemini` (Google OAuth) |
+| OpenCode | `curl -fsSL https://opencode.ai/install \| bash` | `opencode` (configure provider) |
+| Kiro CLI | `brew install --cask kiro-cli` | `kiro-cli login` (AWS Builder ID) |
+
+## Requirements
+
+- Node.js >= 20
+- tmux
+- One of the supported AI coding CLIs (installed and authenticated)
+- Telegram bot token ([@BotFather](https://t.me/BotFather)) or Discord bot token
+- Groq API key (optional, for voice)
+
+> **⚠️** All CLI backends run with `--dangerously-skip-permissions` (or equivalent). See [Security](SECURITY.md).
+
+> **WSL (Windows Subsystem for Linux):** Fully supported. The install script auto-detects WSL and avoids using Windows `node.exe` from PATH. If you encounter PATH issues, add to `/etc/wsl.conf`:
+> ```ini
+> [interop]
+> appendWindowsPath=false
+> ```
+> Then restart WSL (`wsl --shutdown`). Install with: `curl -fsSL https://suzuke.github.io/AgEnD/install.sh | bash`
+
+## Documentation
+
+- [Features](docs/features.md) — detailed feature documentation
+- [CLI Reference](docs/cli.md) — all commands and options
+- [Configuration](docs/configuration.md) — fleet.yaml complete reference
+- [Security](SECURITY.md) — trust model and hardening
+
+## Discord ClassicBot
+
+ClassicBot lets users start AI agents in any Discord text channel using slash commands — no forum topics required.
+
+### Setup
+
+```bash
+# 1. Install Discord plugin
+npm install -g @suzuke/agend-plugin-discord
+
+# 2. Run quickstart (select Discord)
+agend quickstart
+
+# 3. Start the fleet
+agend fleet start
+```
+
+The quickstart will set up both `fleet.yaml` and `classicBot.yaml`. Run `agend quickstart` again to add users or guilds to existing config.
+
+### Usage
+
+| Command | Description |
+|---------|-------------|
+| `/start` | Start an agent in the current channel |
+| `/chat <message>` | Send a message to the agent |
+| `/stop` | Stop the agent in the current channel |
+
+### Server Whitelist
+
+Control which Discord servers can use ClassicBot via `~/.agend/classicBot.yaml`:
+
+```yaml
+defaults:
+  backend: claude-code
+  allowed_guilds:              # Only these servers can /start
+    - "1496855124440780912"
+    - "9876543210123456789"
+```
+
+- **Empty or omitted** `allowed_guilds` → all servers allowed (default)
+- **Primary guild** (fleet.yaml `group_id`) → full access (topic mode + ClassicBot)
+- **Whitelisted guilds** → ClassicBot only (`/start`, `/chat`, `/stop`)
+- **Hot reload** — changes detected every 30 seconds, no restart needed
+
+### Per-Channel Backend
+
+Override the backend for specific channels:
+
+```yaml
+defaults:
+  backend: claude-code
+channels:
+  "1234567890":               # Discord channel ID
+    name: dev-help
+    backend: gemini-cli       # Override for this channel
+```
+
+Backend fallback: channel → `defaults.backend` → `fleet.yaml` defaults → `claude-code`
+
+## Known Limitations
+
+- macOS (launchd) and Linux (systemd) supported; Windows is not
+- Official Telegram plugin in global `enabledPlugins` causes 409 polling conflicts
+- OpenCode and Kiro CLI do not read MCP server `instructions` field — fleet context and workflow templates are not injected into these backends' system prompts. Awaiting upstream fix.
+
+## License
+
+MIT

package/README.zh-TW.md

@@ -0,0 +1,134 @@
+<p align="center">
+  <h1 align="center">AgEnD</h1>
+  <p align="center">
+    <strong>用手機管理一整個 AI coding agent 團隊。</strong>
+  </p>
+  <p align="center">
+    <a href="https://www.npmjs.com/package/@suzuke/agend"><img src="https://img.shields.io/npm/v/@suzuke/agend" alt="npm"></a>
+    <a href="LICENSE"><img src="https://img.shields.io/badge/License-MIT-blue.svg" alt="License: MIT"></a>
+    <a href="https://nodejs.org"><img src="https://img.shields.io/badge/Node.js-%3E%3D%2020-green.svg" alt="Node.js >= 20"></a>
+  </p>
+</p>
+
+AgEnD（**Agent Engineering Daemon**）把你的 Telegram 或 Discord 變成 AI coding agent 的指揮中心。一個 bot，多種 CLI 後端，無限專案 — 每個都是獨立 session，crash 自動恢復，不用顧。
+
+<p align="center">
+  <code>你 → Telegram/Discord → AgEnD → AI Agent 團隊 → 結果回到你的手機</code>
+</p>
+
+[English](README.md) · [功能文件](docs/features.md) · [CLI 參考](docs/cli.md)
+
+---
+
+## 為什麼用 AgEnD？
+
+| 沒有 AgEnD | 有 AgEnD |
+|---|---|
+| 關掉終端機，agent 就斷線 | 系統服務常駐，重開機也不怕 |
+| 一個終端機 = 一個專案 | 一個 bot，無限專案同時跑 |
+| 長時間 session 累積過時 context | 依 max age 自動輪替 session，保持新鮮 |
+| 不知道 agent 半夜在幹嘛 | 每日花費報告 + 卡住偵測通知 |
+| Agent 各做各的，無法協作 | 點對點協作，透過 MCP tools |
+| 無人看管時帳單暴增 | 每個 instance 每日花費上限，自動暫停 |
+
+## 功能亮點
+
+🚀 **Fleet 管理** — 一個 bot、N 個專案。每個 Telegram Forum Topic 就是獨立的 agent session。
+
+🔄 **多後端支援** — Claude Code、Gemini CLI、Codex、OpenCode、Kiro CLI，自由切換或混用。
+
+🤝 **Agent 協作** — Agent 之間透過 MCP tools 互相發現、喚醒、傳訊。General Topic 用自然語言把任務路由到對的 agent。
+
+📱 **手機操控** — 從 Telegram inline 按鈕核准工具使用、重啟 session、管理整個 fleet。
+
+🛡️ **自主又安全** — 花費上限、卡住偵測、model failover、context 輪替，fleet 不用顧也能穩穩跑。
+
+⏰ **持久化排程** — cron 排程任務，SQLite 儲存，重啟不遺失。
+
+🎤 **語音訊息** — 用 Groq Whisper 轉文字，用說的跟 agent 溝通。
+
+📄 **HTML 對話匯出** — 把任何 agent session 匯出成獨立 HTML 檔，方便分享或存檔。
+
+🪞 **Mirror Topic** — 跨 instance 可見性。從另一個 topic 即時觀看其他 agent 的工作。
+
+🖥️ **Web Dashboard** — 瀏覽器即時 fleet 監控，SSE 更新 + 整合聊天介面。
+
+🔌 **可擴充** — Discord adapter、webhook 通知、health endpoint、外部 session 透過 IPC 連入。
+
+## 開始用
+
+一行安裝（macOS / Linux — 自動裝 Node.js（經 nvm）+ tmux + agend，完成後跑 quickstart）：
+
+```bash
+curl -fsSL https://suzuke.github.io/AgEnD/install.sh | bash
+```
+
+或手動安裝：
+
+```bash
+npm install -g @suzuke/agend    # 1. 安裝
+agend quickstart                # 2. 設定 — bot token、backend，搞定
+agend fleet start               # 3. 啟動 fleet 🎉
+```
+
+打開 Telegram，傳訊息給你的 bot，開始用手機寫 code。
+
+## 運作原理
+
+```mermaid
+graph LR
+  You["你<br/>(手機 / 電腦)"] <-->|訊息| Channel["Telegram / Discord<br/>/ Web UI"]
+  Channel <-->|路由| Daemon["AgEnD Daemon"]
+
+  subgraph Fleet
+    Daemon --> General["General<br/>Dispatcher"]
+    Daemon --> A["Instance A<br/>Claude Code<br/>專案 X"]
+    Daemon --> B["Instance B<br/>Gemini CLI<br/>專案 Y"]
+    A <-.->|MCP Tools| B
+    General -.->|路由任務| A
+    General -.->|路由任務| B
+  end
+```
+
+1. **你傳訊息**給 Telegram/Discord bot
+2. 傳到 **General Topic** 的訊息會被解讀並路由到對的 agent。傳到特定 topic 的訊息則直接送到該 instance。
+3. **Agent instance** 在獨立的 tmux session 跑，各有自己的專案和 CLI 後端
+4. **Agent 之間協作** — 透過 MCP tools 委派任務、分享 context、回報結果
+5. **結果回傳**到你的聊天室。權限請求以 inline 按鈕呈現。
+
+## 支援的後端
+
+| Backend | 安裝 | 認證 |
+|---------|------|------|
+| Claude Code | `curl -fsSL https://claude.ai/install.sh \| bash` | `claude`（OAuth）或 `ANTHROPIC_API_KEY` |
+| OpenAI Codex | `npm i -g @openai/codex` | `codex`（ChatGPT 登入）或 `OPENAI_API_KEY` |
+| Gemini CLI | `npm i -g @google/gemini-cli` | `gemini`（Google OAuth） |
+| OpenCode | `curl -fsSL https://opencode.ai/install \| bash` | `opencode`（設定 provider） |
+| Kiro CLI | `brew install --cask kiro-cli` | `kiro-cli login`（AWS Builder ID） |
+
+## 系統需求
+
+- Node.js >= 20
+- tmux
+- 以下任一 AI coding CLI（需安裝並完成認證）
+- Telegram bot token（[@BotFather](https://t.me/BotFather)）或 Discord bot token
+- Groq API key（選用，語音轉文字用）
+
+> **⚠️** 所有 CLI 後端都以 `--dangerously-skip-permissions`（或等效參數）執行。詳見 [Security](SECURITY.md)。
+
+## 文件
+
+- [Features](docs/features.md) — 功能詳細說明
+- [CLI Reference](docs/cli.md) — 所有指令與選項
+- [Configuration](docs/configuration.zh-TW.md) — fleet.yaml 完整設定參考
+- [Security](SECURITY.md) — 信任模型與安全強化
+
+## 已知限制
+
+- 支援 macOS（launchd）和 Linux（systemd），不支援 Windows
+- 全域 `enabledPlugins` 裡有官方 Telegram plugin 會造成 409 polling 衝突
+- OpenCode 和 Kiro CLI 不讀取 MCP server 的 `instructions` 欄位 — fleet context 和 workflow template 不會注入到這些 backend 的 system prompt。等待上游修復。
+
+## 授權
+
+MIT

package/dist/access-path.d.ts

@@ -0,0 +1,10 @@
+/**
+ * Resolve the access.json path for an instance.
+ * Topic mode uses fleet-level access; otherwise per-instance.
+ *
+ * Throws if `instance` is not a safe path segment (per-instance mode only;
+ * topic mode does not embed `instance` in the returned path).
+ */
+export declare function resolveAccessPathFromConfig(dataDir: string, instance: string, fleetChannel: {
+    mode?: string;
+} | undefined): string;

package/dist/access-path.js

@@ -0,0 +1,32 @@
+import { join } from "node:path";
+/**
+ * Conservative whitelist for an instance name when it is going to be used as a
+ * path segment. Allows letters, digits, `_`, `-`, `.`. Empty / pure-`.` /
+ * pure-`..` is rejected so we cannot escape `dataDir/instances/`.
+ *
+ * Defence-in-depth: callers (CLI / fleet config loader) already constrain
+ * instance names, but `resolveAccessPathFromConfig` is invoked from several
+ * entry points and the consequence of a traversal here is reading or writing
+ * an attacker-supplied file path.
+ */
+const VALID_INSTANCE_NAME = /^[A-Za-z0-9._-]+$/;
+function assertSafeInstanceName(instance) {
+    if (!VALID_INSTANCE_NAME.test(instance) || instance === "." || instance === "..") {
+        throw new Error(`Invalid instance name "${instance}" — must match ${VALID_INSTANCE_NAME}`);
+    }
+}
+/**
+ * Resolve the access.json path for an instance.
+ * Topic mode uses fleet-level access; otherwise per-instance.
+ *
+ * Throws if `instance` is not a safe path segment (per-instance mode only;
+ * topic mode does not embed `instance` in the returned path).
+ */
+export function resolveAccessPathFromConfig(dataDir, instance, fleetChannel) {
+    if (fleetChannel?.mode === "topic") {
+        return join(dataDir, "access", "access.json");
+    }
+    assertSafeInstanceName(instance);
+    return join(dataDir, "instances", instance, "access.json");
+}
+//# sourceMappingURL=access-path.js.map

package/dist/access-path.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"access-path.js","sourceRoot":"","sources":["../src/access-path.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAEjC;;;;;;;;;GASG;AACH,MAAM,mBAAmB,GAAG,mBAAmB,CAAC;AAEhD,SAAS,sBAAsB,CAAC,QAAgB;IAC9C,IAAI,CAAC,mBAAmB,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,QAAQ,KAAK,GAAG,IAAI,QAAQ,KAAK,IAAI,EAAE,CAAC;QACjF,MAAM,IAAI,KAAK,CAAC,0BAA0B,QAAQ,kBAAkB,mBAAmB,EAAE,CAAC,CAAC;IAC7F,CAAC;AACH,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,2BAA2B,CACzC,OAAe,EACf,QAAgB,EAChB,YAA2C;IAE3C,IAAI,YAAY,EAAE,IAAI,KAAK,OAAO,EAAE,CAAC;QACnC,OAAO,IAAI,CAAC,OAAO,EAAE,QAAQ,EAAE,aAAa,CAAC,CAAC;IAChD,CAAC;IACD,sBAAsB,CAAC,QAAQ,CAAC,CAAC;IACjC,OAAO,IAAI,CAAC,OAAO,EAAE,WAAW,EAAE,QAAQ,EAAE,aAAa,CAAC,CAAC;AAC7D,CAAC"}

package/dist/adapter-world.d.ts

@@ -0,0 +1,25 @@
+import type { ChannelAdapter, SendOpts, SentMessage } from "./channel/types.js";
+import type { AccessManager } from "./channel/access-manager.js";
+import type { ChannelConfig } from "./types.js";
+/**
+ * AdapterWorld encapsulates a single channel adapter with its associated
+ * access manager and config. Each platform (Telegram, Discord) gets its
+ * own world. Instances are bound to exactly one world.
+ */
+export declare class AdapterWorld {
+    readonly id: string;
+    readonly adapter: ChannelAdapter;
+    readonly accessManager: AccessManager;
+    readonly channelConfig: ChannelConfig;
+    botUsername?: string;
+    botUserId?: string;
+    constructor(id: string, adapter: ChannelAdapter, accessManager: AccessManager, channelConfig: ChannelConfig);
+    get groupId(): string;
+    get type(): string;
+    sendText(chatId: string, text: string, opts?: SendOpts): Promise<SentMessage>;
+    react(chatId: string, messageId: string, emoji: string): Promise<void>;
+    editMessage(chatId: string, messageId: string, text: string): Promise<void>;
+    downloadAttachment(fileId: string): Promise<string>;
+    isAllowed(userId: string | number): boolean;
+    stop(): Promise<void>;
+}

package/dist/adapter-world.js

@@ -0,0 +1,41 @@
+/**
+ * AdapterWorld encapsulates a single channel adapter with its associated
+ * access manager and config. Each platform (Telegram, Discord) gets its
+ * own world. Instances are bound to exactly one world.
+ */
+export class AdapterWorld {
+    id;
+    adapter;
+    accessManager;
+    channelConfig;
+    botUsername;
+    botUserId;
+    constructor(id, adapter, accessManager, channelConfig) {
+        this.id = id;
+        this.adapter = adapter;
+        this.accessManager = accessManager;
+        this.channelConfig = channelConfig;
+    }
+    get groupId() { return String(this.channelConfig.group_id ?? ""); }
+    get type() { return this.channelConfig.type; }
+    // ── Thin wrappers over adapter ──
+    sendText(chatId, text, opts) {
+        return this.adapter.sendText(chatId, text, opts);
+    }
+    react(chatId, messageId, emoji) {
+        return this.adapter.react(chatId, messageId, emoji);
+    }
+    editMessage(chatId, messageId, text) {
+        return this.adapter.editMessage(chatId, messageId, text);
+    }
+    downloadAttachment(fileId) {
+        return this.adapter.downloadAttachment(fileId);
+    }
+    isAllowed(userId) {
+        return this.accessManager.isAllowed(userId);
+    }
+    async stop() {
+        await this.adapter.stop();
+    }
+}
+//# sourceMappingURL=adapter-world.js.map

package/dist/adapter-world.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"adapter-world.js","sourceRoot":"","sources":["../src/adapter-world.ts"],"names":[],"mappings":"AAIA;;;;GAIG;AACH,MAAM,OAAO,YAAY;IAKZ;IACA;IACA;IACA;IAPX,WAAW,CAAU;IACrB,SAAS,CAAU;IAEnB,YACW,EAAU,EACV,OAAuB,EACvB,aAA4B,EAC5B,aAA4B;QAH5B,OAAE,GAAF,EAAE,CAAQ;QACV,YAAO,GAAP,OAAO,CAAgB;QACvB,kBAAa,GAAb,aAAa,CAAe;QAC5B,kBAAa,GAAb,aAAa,CAAe;IACpC,CAAC;IAEJ,IAAI,OAAO,KAAa,OAAO,MAAM,CAAC,IAAI,CAAC,aAAa,CAAC,QAAQ,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;IAC3E,IAAI,IAAI,KAAa,OAAO,IAAI,CAAC,aAAa,CAAC,IAAI,CAAC,CAAC,CAAC;IAEtD,mCAAmC;IAEnC,QAAQ,CAAC,MAAc,EAAE,IAAY,EAAE,IAAe;QACpD,OAAO,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,MAAM,EAAE,IAAI,EAAE,IAAI,CAAC,CAAC;IACnD,CAAC;IAED,KAAK,CAAC,MAAc,EAAE,SAAiB,EAAE,KAAa;QACpD,OAAO,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC,MAAM,EAAE,SAAS,EAAE,KAAK,CAAC,CAAC;IACtD,CAAC;IAED,WAAW,CAAC,MAAc,EAAE,SAAiB,EAAE,IAAY;QACzD,OAAO,IAAI,CAAC,OAAO,CAAC,WAAW,CAAC,MAAM,EAAE,SAAS,EAAE,IAAI,CAAC,CAAC;IAC3D,CAAC;IAED,kBAAkB,CAAC,MAAc;QAC/B,OAAO,IAAI,CAAC,OAAO,CAAC,kBAAkB,CAAC,MAAM,CAAC,CAAC;IACjD,CAAC;IAED,SAAS,CAAC,MAAuB;QAC/B,OAAO,IAAI,CAAC,aAAa,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;IAC9C,CAAC;IAED,KAAK,CAAC,IAAI;QACR,MAAM,IAAI,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC;IAC5B,CAAC;CACF"}

package/dist/agent-cli-instructions.md

@@ -0,0 +1,50 @@
+# AgEnD Agent CLI Reference
+<!-- agend-cli instructions v1 -->
+
+You are an agent managed by AgEnD. Use `agend-agent` commands in bash to communicate.
+All commands output JSON.
+
+## Message Types You Receive
+
+1. **`[user:NAME via telegram] text`** — A human sent you a message.
+   Reply: `agend-agent reply "your response"`
+
+2. **`[from:INSTANCE-NAME] text`** — Another agent sent you a message.
+   Reply: `agend-agent send INSTANCE-NAME "your response"`
+
+3. **`[delegate_task] ...`** — You've been assigned a task.
+   When done: `agend-agent report REQUESTER "summary of results"`
+
+## Quick Reference
+
+```bash
+# Communication
+agend-agent reply "text"                          # Reply to user
+agend-agent send TARGET "message"                 # Message another agent
+agend-agent delegate TARGET "task"                # Assign work
+agend-agent report TARGET "summary"               # Report results
+agend-agent ask TARGET "question"                 # Request info
+agend-agent broadcast "message"                   # Message all agents
+
+# Instance Management
+agend-agent list                                  # List running agents
+agend-agent describe NAME                         # Get agent details
+agend-agent start NAME                            # Start stopped agent
+
+# Task Board
+agend-agent task create "title"                   # Create task
+agend-agent task list                             # List tasks
+agend-agent task claim ID                         # Claim task
+agend-agent task done ID "result"                 # Complete task
+
+# Decisions
+agend-agent decision-post "title" "content"       # Post decision
+agend-agent decision-list                         # List decisions
+```
+
+## Rules
+
+- `[user:... via telegram]` → use `agend-agent reply` (NOT `agend-agent send`)
+- `[from:INSTANCE]` → use `agend-agent send` (NOT `agend-agent reply`)
+- Task flow: `agend-agent delegate` → silent work → `agend-agent report`. Zero messages in between.
+- Never send acknowledgment-only messages. Only send messages with actionable content.

package/dist/agent-cli.d.ts

@@ -0,0 +1,2 @@
+#!/usr/bin/env node
+export {};