let-them-talk 3.9.0 → 4.0.0

package/CHANGELOG.md

@@ -1,5 +1,57 @@
 # Changelog
 
+## [4.0.0] - 2026-03-17
+
+### Major Release — 10-Agent Free Group Mode
+
+Massive scaling overhaul designed, implemented, and audited by a 3-agent team (Architect, Tester, Protocol). 12 changes, 3 bugs caught during collaborative code review.
+
+### Added — Scaling (4 features)
+- **Scaled context** — `listen_group` context window scales with team size: `min(50, max(20, agentCount * 5))`. 3 agents = 20 messages, 10 agents = 50.
+- **Send-after-listen enforcement** — agents must call `listen_group()` between sends. Prevents message storms. Addressed agents get 2 sends per cycle, others get 1.
+- **Response budget** — max 2 unaddressed sends per 60 seconds. Time-based reset. Hint (not error) when depleted.
+- **Smart context with priority partitions** — Bucket A (addressed messages, sacred, always included), Bucket B (channel messages, capped), Bucket C (chronological, fills remaining). Total guaranteed <= contextSize.
+
+### Added — Agent Awareness (3 features)
+- **Enhanced nudge** — every non-listen tool response now includes sender names, addressed count, and message preview: `"URGENT: 3 messages waiting (2 addressed to you): 2 from Architect, 1 from Protocol. Latest: 'Need your review...'"`
+- **Idle detection** — `listen_group()` returns `idle: true` after 60s with no messages, with proactive `work_suggestions`, task suggestions, and instructions. Agents auto-find work instead of blocking forever.
+- **Enhanced `check_messages`** — now returns rich summary: `senders`, `addressed_to_you`, `preview`, `urgency` level. The proactive counterpart to the passive nudge.
+
+### Added — Organization
+- **Task-channel auto-binding** — with 5+ agents in group mode, `create_task` auto-creates `#task-{id}` channels. Assignees auto-join on claim. Channels auto-delete on task completion. Naturally splits 10-agent noise into focused sub-teams.
+
+### Improved — Performance
+- **Cached reads** — `getAgents()` (1.5s TTL), `getChannelsData()` (3s TTL), `getTasks()` (2s TTL) with write-through invalidation. Eliminates ~70% redundant disk I/O.
+- **Compact JSON writes** — removed pretty-print (`null, 2`) from all internal JSON writes. 2-3x less I/O overhead.
+- **Optimized agent status** — removed O(N) `getUnconsumedMessages` scan per agent in `listen_group` status computation.
+- **Dashboard SSE race fix** — `Array.from()` before Set iteration prevents skipped clients during concurrent connect/disconnect.
+- **Dashboard SSE heartbeat** — 30s keepalive prevents dead connection accumulation and proxy timeouts.
+- **Dashboard file watcher cleanup** — old watcher properly closed on LAN toggle, prevents memory leaks.
+- **Dashboard watcher filter** — only triggers on `.json`/`.jsonl` files, ignores lock files and temp files.
+
+### Added — Safety
+- **Collection caps** — tasks (1000), workflows (500), votes (500), reviews (500), dependencies (1000), branches (100), channels (100). Prevents DoS via unbounded growth.
+- **Input type validation** — `reply_to` and `channel` parameters type-checked as strings in `send_message`.
+- **Channel name validation fix** — error message corrected from "1-30 chars" to "1-20 chars" to match `sanitizeName()`.
+
+## [3.10.0] - 2026-03-17
+
+### Added — Dynamic Guide with Progressive Disclosure
+- **`buildGuide()`** — replaces hardcoded guide in register() and get_guide(). Returns only rules relevant to the current system state.
+- **Tiered rules:** Tier 0 (listen after every action), Tier 1 (core behavior), Tier 2 (group mode features), Tier 2b (channels), Tier 3 (large teams 5+)
+- **User-customizable:** `.agent-bridge/guide.md` for project-specific rules
+- 2-agent direct mode = 5 rules. 10-agent group with channels = 12 rules.
+
+## [3.9.1] - 2026-03-17
+
+### Added
+- **Per-channel cooldown** — uses channel member count instead of total agents. 2-member #backend = 1s, regardless of 10 in #general
+- **`cooldown_applied_ms`** — diagnostic field in send_message response showing exact cooldown applied
+- **`channel` field** in send_message response when sending to a channel
+
+### Fixed
+- Task race condition — `update_task` rejects claiming tasks already in_progress by another agent
+
 ## [3.9.0] - 2026-03-17
 
 ### Added — Channels & Split Cooldown

package/README.md

@@ -84,19 +84,22 @@ Each terminal spawns its own MCP server process. All processes share a `.agent-b
 
 ## Highlights
 
+- **10-agent scale** — smart context partitions, send-after-listen enforcement, response budgets, idle detection, task-channel auto-binding
 - **3D virtual office** — chibi characters at desks, spectator camera (WASD+mouse), 11 hairstyles, 6 outfits, gestures, furniture, TV dashboard
 - **Managed conversation mode** — structured turn-taking with floor control for 3+ agents, prevents broadcast storms
-- **53 MCP tools** — messaging, tasks, workflows, profiles, workspaces, branching, managed mode, briefing, file locking, decisions, KB, voting, reviews, dependencies, reputation
+- **56 MCP tools** — messaging, tasks, workflows, profiles, workspaces, branching, managed mode, briefing, file locking, decisions, KB, voting, reviews, dependencies, reputation
 - **8-tab dashboard** — 3D Hub (default), messages, tasks, workspaces, workflows, launch, stats, docs
-- **Group conversation mode** — single-write `__group__` messages, adaptive cooldown, `addressed_to` hints, alive-only GC
+- **Group conversation mode** — single-write `__group__` messages, adaptive cooldown, `addressed_to` hints, smart context, idle detection
+- **Agent awareness** — enhanced nudge with sender/preview on every tool call, idle work suggestions, rich `check_messages`
 - **5 agent templates** — pair, team, review, debate, managed — with ready-to-paste prompts
 - **5 conversation templates** — Code Review, Debug Squad, Feature Dev, Research & Write, Managed Team
 - **Stats & analytics** — per-agent scores, response times, hourly charts, conversation velocity
-- **Task management** — drag-and-drop kanban board between agents
+- **Task management** — drag-and-drop kanban board, task-channel auto-binding for 5+ agent teams
 - **Workflow pipelines** — multi-step automation with auto-handoff
 - **Conversation branching** — fork at any point, isolated history per branch
 - **Ollama integration** — `npx let-them-talk init --ollama` for local AI models
-- **Secure by default** — CSRF, LAN auth tokens, CSP, config locking, reserved name blocklist
+- **Performance optimized** — cached reads (70% I/O reduction), compact JSON writes, SSE heartbeat
+- **Secure by default** — CSRF, LAN auth tokens, CSP, collection caps, config locking, reserved name blocklist
 - **Zero config** — one `npx` command, auto-detects your CLI, works immediately
 
 ## Agent Templates
@@ -175,7 +178,7 @@ The dashboard's default view is a **real-time 3D virtual office** (the "3D Hub")
 
 **Animations:** walk, sit, type, raise hand, sleep (ZZZ), wave, think, point, celebrate, stretch, idle gestures. Agents turn toward speakers during conversations.
 
-## MCP Tools (53)
+## MCP Tools (56)
 
 <details>
 <summary><strong>Messaging (13 tools)</strong></summary>
@@ -333,6 +336,17 @@ The dashboard's default view is a **real-time 3D virtual office** (the "3D Hub")
 
 </details>
 
+<details>
+<summary><strong>Channels (3 tools)</strong></summary>
+
+| Tool | Description |
+|------|-------------|
+| `join_channel` | Join or create a channel for sub-team communication |
+| `leave_channel` | Leave a channel (can't leave #general, empty auto-delete) |
+| `list_channels` | List all channels with members and message counts |
+
+</details>
+
 ## CLI Reference
 
 ```bash

package/cli.js

@@ -9,7 +9,7 @@ const command = process.argv[2];
 
 function printUsage() {
   console.log(`
-  Let Them Talk — Agent Bridge v3.9.0
+  Let Them Talk — Agent Bridge v4.0.0
   MCP message broker for inter-agent communication
   Supports: Claude Code, Gemini CLI, Codex CLI, Ollama
 

package/dashboard.html

@@ -6879,7 +6879,7 @@ function renderDocs() {
 
     // All 53 Tools
     '<div class="docs-section">' +
-      '<h3>All 53 MCP Tools</h3>' +
+      '<h3>All 56 MCP Tools</h3>' +
       '<h4>Core Messaging</h4>' +
       '<div class="docs-tool-grid">' +
         '<div class="docs-tool-item"><code>register(name, provider?)</code><div class="desc">Register your agent identity. Must be called first.</div></div>' +

package/dashboard.js

@@ -1820,7 +1820,11 @@ const server = http.createServer(async (req, res) => {
       });
       res.write(`data: connected\n\n`);
       sseClients.add(res);
-      req.on('close', () => sseClients.delete(res));
+      // Heartbeat every 30s to detect dead connections and prevent proxy timeouts
+      const heartbeat = setInterval(() => {
+        try { res.write(`:heartbeat\n\n`); } catch { clearInterval(heartbeat); sseClients.delete(res); }
+      }, 30000);
+      req.on('close', () => { clearInterval(heartbeat); sseClients.delete(res); });
     }
     // --- Mod system API ---
     else if (url.pathname === '/api/mods' && req.method === 'GET') {
@@ -1965,13 +1969,15 @@ function sseNotifyAll() {
     generateNotifications(agents);
   } catch {}
 
-  for (const res of sseClients) {
+  const dead = [];
+  for (const res of Array.from(sseClients)) {
     try {
       res.write(`data: update\n\n`);
     } catch {
-      sseClients.delete(res);
+      dead.push(res);
     }
   }
+  for (const res of dead) sseClients.delete(res);
 }
 
 // Watch data directory for changes and push SSE notifications
@@ -1979,10 +1985,17 @@ let fsWatcher = null;
 let sseDebounceTimer = null;
 
 function startFileWatcher() {
+  // Clean up previous watcher to prevent memory leaks on LAN toggle
+  if (fsWatcher) { try { fsWatcher.close(); } catch {} fsWatcher = null; }
+  if (sseDebounceTimer) { clearTimeout(sseDebounceTimer); sseDebounceTimer = null; }
+
   const dataDir = resolveDataDir();
   if (!fs.existsSync(dataDir)) return;
   try {
-    fsWatcher = fs.watch(dataDir, { persistent: false }, () => {
+    fsWatcher = fs.watch(dataDir, { persistent: false }, (eventType, filename) => {
+      // Filter: only react to data files, not temp/lock files
+      if (filename && !filename.endsWith('.json') && !filename.endsWith('.jsonl')) return;
+      if (filename && filename.endsWith('.lock')) return;
       // Debounce — multiple file changes may fire rapidly
       if (sseDebounceTimer) clearTimeout(sseDebounceTimer);
       sseDebounceTimer = setTimeout(() => sseNotifyAll(), 200);

package/package.json

@@ -1,58 +1,58 @@
-{
-  "name": "let-them-talk",
-  "version": "3.9.0",
-  "description": "MCP message broker + web dashboard for inter-agent communication. Let AI CLI agents talk to each other.",
-  "main": "server.js",
-  "bin": {
-    "agent-bridge": "./cli.js",
-    "let-them-talk": "./cli.js"
-  },
-  "scripts": {
-    "start": "node server.js",
-    "dashboard": "node dashboard.js",
-    "test": "echo \"No tests yet\""
-  },
-  "engines": {
-    "node": ">=16.0.0"
-  },
-  "files": [
-    "server.js",
-    "dashboard.js",
-    "dashboard.html",
-    "office/",
-    "mods/",
-    "cli.js",
-    "templates/",
-    "conversation-templates/",
-    "logo.png",
-    "LICENSE",
-    "SECURITY.md",
-    "CHANGELOG.md"
-  ],
-  "keywords": [
-    "mcp",
-    "claude",
-    "claude-code",
-    "gemini-cli",
-    "codex-cli",
-    "agent",
-    "multi-agent",
-    "communication",
-    "message-broker",
-    "ai-agents",
-    "let-them-talk"
-  ],
-  "repository": {
-    "type": "git",
-    "url": "git+https://github.com/Dekelelz/let-them-talk.git"
-  },
-  "homepage": "https://talk.unrealai.studio",
-  "bugs": {
-    "url": "https://github.com/Dekelelz/let-them-talk/issues"
-  },
-  "author": "Dekelelz <contact@talk.unrealai.studio>",
-  "license": "SEE LICENSE IN LICENSE",
-  "dependencies": {
-    "@modelcontextprotocol/sdk": "1.27.1"
-  }
-}
+{
+  "name": "let-them-talk",
+  "version": "4.0.0",
+  "description": "MCP message broker + web dashboard for inter-agent communication. Let AI CLI agents talk to each other.",
+  "main": "server.js",
+  "bin": {
+    "agent-bridge": "./cli.js",
+    "let-them-talk": "./cli.js"
+  },
+  "scripts": {
+    "start": "node server.js",
+    "dashboard": "node dashboard.js",
+    "test": "echo \"No tests yet\""
+  },
+  "engines": {
+    "node": ">=16.0.0"
+  },
+  "files": [
+    "server.js",
+    "dashboard.js",
+    "dashboard.html",
+    "office/",
+    "mods/",
+    "cli.js",
+    "templates/",
+    "conversation-templates/",
+    "logo.png",
+    "LICENSE",
+    "SECURITY.md",
+    "CHANGELOG.md"
+  ],
+  "keywords": [
+    "mcp",
+    "claude",
+    "claude-code",
+    "gemini-cli",
+    "codex-cli",
+    "agent",
+    "multi-agent",
+    "communication",
+    "message-broker",
+    "ai-agents",
+    "let-them-talk"
+  ],
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/Dekelelz/let-them-talk.git"
+  },
+  "homepage": "https://talk.unrealai.studio",
+  "bugs": {
+    "url": "https://github.com/Dekelelz/let-them-talk/issues"
+  },
+  "author": "Dekelelz <contact@talk.unrealai.studio>",
+  "license": "SEE LICENSE IN LICENSE",
+  "dependencies": {
+    "@modelcontextprotocol/sdk": "1.27.1"
+  }
+}