mcp-swarm 1.1.2 → 1.1.3

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/CHANGELOG.md CHANGED
@@ -9,143 +9,39 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
9
9
 
10
10
  ---
11
11
 
12
- ## [1.1.2] - 2026-02-09
12
+ ## [1.1.3] - 2026-02-09
13
13
 
14
- ### 🏗️ Code Quality & Security
14
+ ### What's New
15
15
 
16
- #### Added
16
+ #### 📱 Telegram Bot Integration
17
+ - **Full Telegram notifications** — Task events, agent status, CI errors, code reviews — all delivered to your Telegram.
18
+ - **Bilingual setup guide** — Complete `TELEGRAM.md` with step-by-step instructions in English and Russian.
19
+ - **@userinfobot support** — Easy way to discover your Telegram User ID.
20
+ - **Bot commands** — `/start`, `/projects`, `/status`, `/agents`, `/tasks`, `/myid`, `/reviews`, `/approve`, `/reject`.
21
+ - **Environment variables** — `TELEGRAM_USER_ID` and `TELEGRAM_BOT_URL` for all MCP configurations.
17
22
 
18
- - **ESLint + Prettier** Full linting and formatting setup with `typescript-eslint`. New npm scripts: `lint`, `lint:fix`, `format`, `format:check`.
19
- - **fs-sandbox** — File system sandbox module (`src/fsSandbox.ts`) restricts agent file operations to the project boundary, preventing path-traversal attacks.
23
+ #### 🏗️ Code Quality & Security
24
+ - **ESLint + Prettier** — Full linting and formatting setup with `typescript-eslint`. Scripts: `lint`, `lint:fix`, `format`, `format:check`.
25
+ - **fs-sandbox** — File system sandbox (`src/fsSandbox.ts`) prevents path-traversal attacks by restricting agent file operations to the project boundary.
26
+ - **Dashboard refactoring** — Extracted 133-line inline HTML from `companion.ts` into `dashboard.ts` module.
20
27
 
21
- #### Changed
28
+ #### 🔭 Observability & Control
29
+ - **File Logging** — Companion logs to `~/.mcp-swarm/logs/companion-YYYY-MM-DD.log` with 7-day rotation.
30
+ - **`mcp-swarm-doctor`** — CLI diagnostics: Node.js, Git, companion status, ports, logs, Hub URL, IDE configs.
31
+ - **Interactive Dashboard** — Pause/Resume/Shutdown buttons + Toast notifications at `http://localhost:37373`.
32
+ - **Auto-Update Notifier** — Warns on startup if a newer npm version is available.
22
33
 
23
- - **Dashboard refactoring** — Extracted 133-line inline HTML from `companion.ts` into dedicated `dashboard.ts` module. Companion is now 700 lines (was 820).
24
- - **ESLint config** Flat config (`eslint.config.js`) with TypeScript support, ignoring `dist/`, `node_modules/`, `dashboard/`, `cloudflare/`.
25
- - **Prettier config** — `.prettierrc` with 120-char line width, double quotes, trailing commas, LF line endings.
34
+ #### 🐝 Web Dashboard
35
+ - **Dark-themed dashboard** at `http://localhost:37373` with auto-refresh every 5s.
36
+ - **PID file** + **Graceful shutdown** `~/.mcp-swarm/companion.pid` with SIGTERM/SIGINT handlers.
37
+ - **`/health` endpoint** — `{ ok, pid, uptime }` for monitoring.
38
+ - **Unit tests** — Tests for `normalizeGitRemote` and PID file management.
26
39
 
27
40
  ---
28
41
 
29
- ## [1.1.1] - 2026-02-09
42
+ ### Configuration
30
43
 
31
- ### 🔭 Observability & Control
32
-
33
- #### Added
34
-
35
- - **File Logging** — All companion logs written to `~/.mcp-swarm/logs/companion-YYYY-MM-DD.log` with automatic 7-day rotation.
36
- - **`mcp-swarm-doctor`** — New CLI self-diagnostics tool. Checks Node.js, Git, companion status, port, logs, Hub URL, IDE configs, and npm version.
37
- - **Interactive Dashboard** — Buttons for Pause/Resume/Shutdown/Copy Project ID directly in the web dashboard at `http://localhost:37373`. Toast notifications for action feedback.
38
- - **Auto-Update Notifier** — On startup, companion checks npm registry and warns if a newer version is available.
39
-
40
- #### Changed
41
-
42
- - **Graceful shutdown** now properly closes file log streams before exit.
43
- - **Dashboard footer** updated to v1.1.
44
-
45
- ---
46
-
47
- ## [1.1.0] - 2026-02-09
48
-
49
- ### 🐝 Web Dashboard + DX Improvements
50
-
51
- #### Added
52
-
53
- - **Web Dashboard** — Beautiful dark-themed dashboard at `http://localhost:37373` with auto-refresh every 5s. Shows agent name, role, bridge status, uptime, PID, and all API endpoints.
54
- - **PID file** — Companion now writes `~/.mcp-swarm/companion.pid` on startup and removes it on exit for process discovery.
55
- - **Graceful shutdown** — SIGTERM/SIGINT handlers with PID cleanup.
56
- - **`/health` endpoint** — Returns JSON with `{ ok, pid, uptime }` for monitoring.
57
- - **`examples/`** — Ready-to-use configs for Claude Code, Cursor, Windsurf, OpenCode, and local mode.
58
- - **Issue Templates** — Bug report, feature request, and question templates in `.github/ISSUE_TEMPLATE/`.
59
- - **Unit tests** — Tests for `normalizeGitRemote` and PID file management (`src/tests/projectId.test.ts`).
60
- - **New badges** — Node.js 18+, TypeScript 5.0, MCP Protocol added to README EN + RU.
61
- - **Quick Demo** — ASCII workflow diagram added to README EN + RU showing multi-agent coordination.
62
-
63
- #### Changed
64
-
65
- - **`smartTools.ts`** — Legacy monolith renamed to `smartTools.legacy.ts` (unused, server uses modular `smartTools/`).
66
- - **`--version` flag** — Added to `mcp-swarm-remote` CLI (`-v`, `-V`, `--version`).
67
- - **`readFileSync` import** — Added to `remote/index.ts` for version reading.
68
-
69
- ---
70
-
71
- ## [1.0.11] - 2026-02-09
72
-
73
- ### 🎨 Flag Badges & Community CTA
74
-
75
- #### Changed
76
-
77
- - **Flag badges** — Replaced Unicode country flag emoji (🇷🇺/🇬🇧) with `<img>` badges from flagcdn.com. Flags now display correctly on npm, GitHub, and all platforms.
78
- - **Community CTA section** — Added "⭐ Support the Project" section to both README (EN + RU) with links to Issues, Discussions, and Fork & PR.
79
-
80
- ---
81
-
82
- ## [1.0.10] - 2026-02-09
83
-
84
- ### 🔄 CI/CD Auto-Publish & Docs Update
85
-
86
- #### Changed
87
-
88
- - **CI/CD fully automated** — GitHub Actions now auto-publishes to npm AND creates GitHub Release when a new version is pushed to `main`. Skips gracefully if version already exists.
89
- - **README updated** — Both README.md and README.ru.md updated from v1.0.7 → v1.0.10 with current feature descriptions.
90
-
91
- ---
92
-
93
- ## [1.0.9] - 2026-02-09
94
-
95
- ### 🧭 Smart Project ID with Git Init Suggestions
96
-
97
- #### Added
98
-
99
- - **Git init suggestions** — When a project directory has no git repo, companion now prints step-by-step instructions: `git init` → `git add -A` → `git commit` → `gh repo create`.
100
- - **Remote-only suggestions** — When git exists but no remote origin, suggests `gh repo create` or `git remote add origin`.
101
- - **`ProjectIdResult.suggestions`** — New `suggestions[]` field in project ID resolution result, available to tools and agents.
102
-
103
- #### Changed
104
-
105
- - **Clean folder-based IDs** — Path-based project IDs no longer include hash suffix. `Intop Saas` → `intop_saas` (was `intop_saas_a1b2c3`).
106
- - **`isGitInitialized()` and `hasRemoteOrigin()`** — New helper functions for granular git status detection.
107
-
108
- ---
109
-
110
- ## [1.0.8] - 2026-02-09
111
-
112
- ### 🔧 Companion Bridge Auto-Restart
113
-
114
- #### Fixed
115
-
116
- - **Critical: Bridge auto-restart** — If companion is running but bridge is disconnected (e.g. old companion without `MCP_SERVER_URL`), `ensureCompanion()` now kills the old process and restarts with correct environment variables.
117
- - **`isBridgeConnected()`** — New function to verify bridge WebSocket status via companion control API.
118
- - **`killCompanion()`** — New function to gracefully shut down companion before restart.
119
-
120
- #### Changed
121
-
122
- - **ensureCompanion() logic rewritten:**
123
- 1. Check if companion process is running
124
- 2. Check if bridge is actually connected
125
- 3. If running but disconnected → kill + restart with `MCP_SERVER_URL`
126
- 4. If not running → start fresh
127
-
128
- ---
129
-
130
- ## [1.0.7] - 2026-02-09
131
-
132
- ### 🌉 Full Bridge Coverage — All 26 Smart Tools via Remote
133
-
134
- #### Fixed
135
-
136
- - **Critical: Bridge auto-start** — `mcp-swarm-remote` now passes `MCP_SERVER_URL` to companion daemon. Previously the companion couldn't know where to connect the bridge, so all remote tool calls returned `{ bridgeConnected: false }`.
137
- - **Documentation: Full startup flow** — README now includes complete configuration examples for both Remote and Local modes with `SWARM_HUB_URL`, and step-by-step explanation of what happens at startup.
138
-
139
- #### Changed
140
-
141
- - **Universal bridge delegation** — `bridge.ts` now imports `allSmartTools` handlers and delegates ALL tool calls through them instead of manually implementing 3 tools with limited actions.
142
- - Before: only `swarm_file` (read/write/list), `swarm_git` (status/sync), `swarm_agent` (register/whoami) worked through bridge
143
- - After: all 26 tools × all actions work through bridge (swarm_task, swarm_plan, swarm_quality, swarm_vector, etc.)
144
- - **Simplified tool routing** — `toolNeedsBridge()` on Cloudflare Worker simplified from 21-line selective logic to `toolName.startsWith("swarm_")` — routes ALL swarm tools through bridge.
145
-
146
- #### Configuration
147
-
148
- **Option A: Remote (recommended)**
44
+ **Option A: Remote (Recommended)**
149
45
 
150
46
  ```json
151
47
  {
@@ -158,7 +54,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
158
54
  "--url", "https://mcp-swarm-server.YOUR-SUBDOMAIN.workers.dev/mcp"
159
55
  ],
160
56
  "env": {
161
- "SWARM_HUB_URL": "wss://mcp-swarm-hub.YOUR-SUBDOMAIN.workers.dev/ws"
57
+ "SWARM_HUB_URL": "wss://mcp-swarm-hub.YOUR-SUBDOMAIN.workers.dev/ws",
58
+ "TELEGRAM_USER_ID": "YOUR_TELEGRAM_USER_ID",
59
+ "TELEGRAM_BOT_URL": "https://YOUR-TELEGRAM-BOT.workers.dev"
162
60
  }
163
61
  }
164
62
  }
@@ -174,1052 +72,61 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
174
72
  "command": "node",
175
73
  "args": ["C:/path/to/Swarm_MCP/dist/serverSmart.js"],
176
74
  "env": {
177
- "SWARM_HUB_URL": "wss://mcp-swarm-hub.YOUR-SUBDOMAIN.workers.dev/ws"
75
+ "SWARM_HUB_URL": "wss://mcp-swarm-hub.YOUR-SUBDOMAIN.workers.dev/ws",
76
+ "TELEGRAM_USER_ID": "YOUR_TELEGRAM_USER_ID",
77
+ "TELEGRAM_BOT_URL": "https://YOUR-TELEGRAM-BOT.workers.dev"
178
78
  }
179
79
  }
180
80
  }
181
81
  }
182
82
  ```
183
83
 
184
- #### What Happens at Startup (Remote)
185
-
186
- ```
187
- 1. npx downloads mcp-swarm@latest from npm
188
- 2. mcp-swarm-remote starts checks if companion is running
189
- 3. If not → starts companion with:
190
- • MCP_SERVER_URL (from --url) → Bridge auto-connects to your Worker
191
- • SWARM_HUB_URL (from env) → WebSocket to Hub for coordination
192
- 4. Companion starts:
193
- • Bridge → WebSocket → MCP Server Worker (executes 26 tools locally)
194
- • Hub → WebSocket → Hub Worker (real-time agent sync)
195
- 5. All 26 smart tools work! ✅
196
- ```
197
-
198
- ---
199
-
200
- ## [0.9.19] - 2026-02-08
201
-
202
- ### 🚀 Major Release: Smart Routing, Memory, Agent Teams, MCP Bridges
203
-
204
- #### Added
205
-
206
- - **Smart Router Proxy** (`smartRouterProxy.ts`) — Cost optimization engine inspired by distiq-code
207
- - Automatic model tier downgrade (Opus → Sonnet when safe)
208
- - Semantic cache for repeated/similar requests (SHA-256 + similarity)
209
- - Prompt caching suggestions (cache_control breakpoints for Anthropic API)
210
- - Request classification: complexity analysis, token estimation, tier recommendation
211
- - Real-time cost savings tracking and statistics
212
-
213
- - **Swarm Memory** (`swarmMemory.ts`) — Hybrid memory system (claude-mem + claude-cognitive)
214
- - 3-tier Context Router: hot (current session) / warm (24h) / cold (archive)
215
- - Pool Coordinator for multi-agent memory synchronization
216
- - Lifecycle hooks: session_start, prompt_submit, response_ready, session_end
217
- - 3-layer search: keyword/tag → timeline → full observations
218
- - Auto-compression for older entries (context window savings)
219
-
220
- - **MCP Linear Bridge** (`mcpLinearBridge.ts`) — Auto-sync swarm tasks ↔ Linear issues
221
- - Status mapping: open → Todo, in_progress → In Progress, done → Done
222
- - Task registration and bidirectional sync
223
- - Passive activation (only when mcp-linear MCP is detected)
224
-
225
- - **MCP Context7 Bridge** (`mcpContext7Bridge.ts`) — Up-to-date documentation integration
226
- - Auto-detect project tech stack (React, Next.js, Express, Prisma, Supabase, Tailwind...)
227
- - Documentation cache with configurable TTL (default 24h)
228
- - Lookup with cache-first strategy
229
-
230
- - **Claude-Flow Bridge** (`claudeFlowBridge.ts`) — Skills routing and RAG pipeline
231
- - Q-learning based skill routing (epsilon-greedy exploration/exploitation)
232
- - Skill registry with quality tracking (Q-values, success rates)
233
- - RAG configuration for vector search → context injection
234
-
235
- - **Agent Teams** (`agentTeams.ts`) — Multi-agent team coordination
236
- - Team creation with roles: lead, developer, reviewer, tester, specialist
237
- - Task delegation with dependency tracking
238
- - Auto-rebalancing: redistribute tasks when agent goes offline
239
- - Team-level broadcast messaging
240
- - RAC (Retrieval Augmented Coding) search placeholder
241
-
242
- - **Skills Discovery** (`skillsDiscovery.ts`) — IDE-agnostic skill detection
243
- - Scans 6+ IDE formats: Gemini, Antigravity, Claude, Cursor, Codex, Windsurf
244
- - Normalizes skills to unified format
245
- - Task-based skill recommendation (keyword scoring)
246
- - Cross-IDE skill import/export
247
-
248
- ## [0.9.18] - 2026-02-08
249
-
250
- ### 🧠 Интеграция новых AI-моделей
251
-
252
- #### Добавлено
253
-
254
- - **Claude Opus 4.6** — новый флагман Anthropic в MoE Router
255
- - 1M токенов контекст (первый Opus с миллионным контекстом)
256
- - Adaptive thinking — модель сама решает когда использовать extended thinking
257
- - Context compaction — автоматическое сжатие при приближении к лимиту
258
- - 128K output tokens — большие ответы без разбиения
259
- - Agent Teams — координация нескольких агентов через tmux
260
- - $5/$25 per MTok (≤200K), $10/$37.50 (>200K)
261
-
262
- - **GPT-5.3 Codex** — новый флагман OpenAI для агентного кодинга
263
- - 256K контекст
264
- - Оптимизирован для code_generation, debugging, reasoning
265
- - ~$2/$15 per MTok (estimated)
266
-
267
- - **Kimi K2.5** — premium модель Moonshot AI
268
- - 128K контекст, фокус на code_generation, code_review, debugging
269
- - ~$1/$5 per MTok (estimated)
270
-
271
- #### Изменено
272
-
273
- - `ModelProvider` расширен на `"moonshot"` (Kimi/Moonshot AI)
274
- - Claude Opus 4.5 понижен с `flagship` → `premium` (заменён Opus 4.6)
275
- - MoE Router: **19 моделей** (было 16) — 4 Anthropic, 9 OpenAI, 5 Google, 1 Moonshot
276
-
277
- ---
278
-
279
- ## [0.9.17] - 2026-02-08
280
-
281
- ### 🏗️ Smart Tools Modularization
282
-
283
- #### Changed
284
-
285
- - **`src/smartTools.ts`** — split into **9 modules** in `src/smartTools/`:
286
- - `core.ts` — swarm_agent, swarm_control, swarm_pulse, swarm_companion
287
- - `tasks.ts` — swarm_task, swarm_plan, swarm_briefing, swarm_spec
288
- - `files.ts` — swarm_file, swarm_worktree, swarm_snapshot
289
- - `git.ts` — swarm_git, swarm_hooks, swarm_dependency
290
- - `collaboration.ts` — swarm_chat, swarm_message, swarm_review, swarm_voting, swarm_auction, swarm_brainstorm
291
- - `security.ts` — swarm_defence, swarm_consensus, swarm_mcp
292
- - `analytics.ts` — swarm_cost, swarm_quality, swarm_regression, swarm_session
293
- - `intelligence.ts` — swarm_sona, swarm_moe, swarm_vector, swarm_booster, swarm_context_pool, swarm_context
294
- - `infra.ts` — swarm_health, swarm_immune, swarm_external, swarm_platform, swarm_knowledge, etc.
295
- - **`src/smartTools/index.ts`** — centralized re-export of all modules
296
- - **`src/serverSmart.ts`** — updated to import from `smartTools/index.js`
297
-
298
- ### 🔒 Cloudflare Workers with Authentication
299
-
300
- #### Added
301
-
302
- - **`cloudflare/abdr-hub/`** — new Cloudflare Worker `abdr-swarm-hub`
303
- - Bearer Token + Query parameter authentication
304
- - WebSocket bridge for agents
305
- - REST API for swarm management
306
- - **`cloudflare/abdr-server/`** — new Cloudflare Worker `abdr-swarm-server`
307
- - Streamable HTTP Transport (MCP spec 2025-03-26)
308
- - All 54 Smart Tools
309
- - Automatic Bearer Token forwarding to Hub
310
- - **`cloudflare/telegram-bot/`** → renamed to `abdr-swarm-telegram`
311
- - `SWARM_AUTH_TOKEN` protection for `/register` endpoint
312
- - Bearer Token for Hub API calls
313
-
314
- ### 🔄 CI/CD Pipeline
315
-
316
- #### Changed
317
-
318
- - **`.github/workflows/ci.yml`** — improvements:
319
- - Triggers for `main` and `develop` branches
320
- - TypeScript check (`tsc --noEmit`) before build
321
- - Node.js matrix: 18, 20, 22
322
- - Automatic npm publish on push to `main`
323
-
324
- ### 🗑️ Legacy Code Removed
325
-
326
- #### Removed
327
-
328
- - **`src/tools.ts`** — old 41 tools (replaced by `smartTools/`)
329
- - **`src/server.ts`** — old server (replaced by `serverSmart.ts`)
330
-
331
- ### 📊 Dashboard
332
-
333
- #### Changed
334
-
335
- - Updated version in footer: v0.9.12 → v0.9.17
336
-
337
- ### 📦 Updates
338
-
339
- - `package.json` → version `0.9.17`
340
-
341
- ### 🔄 How to Update
342
-
343
- ```bash
344
- npm install -g mcp-swarm@latest
345
- # or
346
- npm update mcp-swarm
347
- ```
348
-
349
- ---
350
-
351
- ## [0.9.16] - 2026-02-08
352
-
353
- ### 🐛 Critical Fix: Zod Compatibility
354
-
355
- **BREAKING FIX:** Fixed `keyValidator._parse is not a function` error that made all 54 Smart Tools completely non-functional in Antigravity and similar clients.
356
-
357
- #### Root Cause
358
- - `zod@^3.23.8` in `package.json` auto-installed `3.25.76` — essentially a Zod v4 bridge version
359
- - Zod v4 removed the internal `_parse()` method used by clients for JSON Schema validation
360
- - Result: **no Smart Tool worked** due to input schema parsing errors
361
-
362
- #### Fix
363
- - `zod` pinned to exact version `3.23.8` (pure Zod v3, no v4 bridge)
364
- - Prevents automatic upgrades to incompatible versions
365
-
366
- > 🙏 We apologize for the inconvenience. This bug affected all users who updated via npm.
367
-
368
- ### 🔒 Security: Cloudflare Workers
369
-
370
- - Removed deployed workers `mcp-swarm-hub` and `mcp-swarm-server` (URLs were exposed in commits)
371
- - Worker sources **preserved** in `cloudflare/` — will be redeployed with new names
372
-
373
- ### 📦 Updates
374
-
375
- - `serverSmart.ts` — dynamic version from `package.json` instead of hardcode
376
- - `server.ts` — marked as deprecated (`@ts-nocheck`), use `serverSmart.ts`
377
- - `.gitignore` — added patterns for test files
378
-
379
- ### 🔄 How to Update
380
-
381
- ```bash
382
- npm install -g mcp-swarm@latest
383
- # or
384
- npm update mcp-swarm
385
- ```
386
-
387
- ---
388
-
389
- ## [0.9.15] - 2026-02-08
84
+ | Variable | Required | Description |
85
+ |----------|----------|-------------|
86
+ | `SWARM_HUB_URL` | ✅ | WebSocket URL of your deployed Hub worker |
87
+ | `TELEGRAM_USER_ID` | Optional | Your Telegram User ID (get it via [@userinfobot](https://t.me/userinfobot)) |
88
+ | `TELEGRAM_BOT_URL` | Optional | URL of your deployed Telegram bot worker |
390
89
 
391
- ### 📖 Bilingual Documentation
392
-
393
- #### Changed
394
-
395
- - **`README.md`** — fully translated to **English** (~1100 lines)
396
- - Professional translation of all sections: tools, installation, configuration, architecture
397
- - Added link to Russian version: `🇷🇺 Читать на русском`
398
- - **`package.json`** — description updated to English
399
- - **Version updated in heading** — v0.9.15
400
-
401
- #### Added
402
-
403
- - **`README.ru.md`** — Russian version of README created
404
- - Full original README content in Russian
405
- - Link to English version: `🇬🇧 Read in English`
406
-
407
- ---
408
-
409
- ## [0.9.14] - 2026-02-08
410
-
411
- ### 🐛 npx Command Fix
412
-
413
- #### Fixed
414
-
415
- - **`README.md`** — fixed `npx mcp-swarm-remote` commands:
416
- - Before: `npx mcp-swarm-remote --url ...` (E404 error, package not found)
417
- - After: `npx -y -p mcp-swarm mcp-swarm-remote --url ...`
418
- - **`REMOTE.md`** — same fix applied to all IDE configuration examples
419
- - **Version updated** — v0.9.14
420
-
421
- #### Root Cause
422
-
423
- - `npx mcp-swarm-remote` tried to find a **separate** package `mcp-swarm-remote` on npm
424
- - `mcp-swarm-remote` is a binary within the `mcp-swarm` package, so the `-p mcp-swarm` flag is required
425
-
426
- ---
427
-
428
- ## [0.9.13] - 2026-02-05
429
-
430
- ### Self-Hosted Infrastructure
431
-
432
- **BREAKING CHANGE:** Removed hardcoded public server URLs. Users must now deploy their own Cloudflare Workers.
433
-
434
- #### Changed
435
-
436
- - **README.md** — Complete rewrite of installation section
437
- - Added Cloudflare Free Tier explanation and limits
438
- - Step-by-step deployment guide for Hub, MCP Server, Telegram Bot
439
- - Explained what YOUR-SUBDOMAIN means
440
-
441
- - **wrangler.toml files** — All 3 Workers now have placeholders
442
- - `cloudflare/hub/wrangler.toml` — Added deployment instructions
443
- - `cloudflare/mcp-server/wrangler.toml` — HUB_URL placeholder
444
- - `cloudflare/telegram-bot/wrangler.toml` — SWARM_HUB_URL placeholder
445
-
446
- - **install.ps1 / install.sh** — Now ask for user's server URLs
447
- - No more hardcoded URLs
448
- - Instructions for deploying first
449
-
450
- - **mcp-swarm-remote** — `--url` is now **required**
451
- - Shows helpful error message with deployment link
452
- - Added `--help` option
453
-
454
- - **Dashboard** — `.env.example` uses placeholder
455
- - `useWebSocket.ts` warns if HUB_URL not configured
456
-
457
- - **Agent rules** (AGENTS.md, CLAUDE.md, GEMINI.md, AGENT.md)
458
- - Updated with YOUR-SUBDOMAIN placeholders
459
-
460
- #### Why This Change?
461
-
462
- - **Privacy**: Your data stays on your infrastructure
463
- - **No shared limits**: Full Cloudflare Free Tier for yourself
464
- - **Customization**: Modify Workers as needed
465
- - **Transparency**: No hidden public server
90
+ > 📱 See [TELEGRAM.md](./TELEGRAM.md) for full Telegram setup instructions.
466
91
 
467
92
  ---
468
93
 
469
- ## [0.9.12] - 2026-02-05
470
-
471
- ### Dashboard Real-time Updates
472
-
473
- #### Added
474
-
475
- - **WebSocket Hook** (`dashboard/src/hooks/useWebSocket.ts`)
476
- - Real-time connection to Swarm Hub
477
- - Auto-reconnect with exponential backoff
478
- - Keep-alive ping every 25 seconds
479
- - Event filtering for specific event types
94
+ ### Platform Highlights
480
95
 
481
- - **New Core Widgets** (`dashboard/src/components/widgets/CoreWidgets.tsx`)
482
- - **ConnectionStatusWidget** — Shows Hub connection status with reconnect button
483
- - **ActivityTimelineWidget** — Real-time event stream from Hub
484
- - **FileLocksWidget** — Active file reservations with live updates
485
- - **CostTrackingWidget** — API usage and budget progress bars
486
- - **VotingWidget** — Distributed consensus proposals and votes
96
+ These are the key capabilities built into MCP Swarm across all versions:
487
97
 
488
- - **Live Indicators**
489
- - Green pulsing dot for connected/active status
490
- - LIVE/OFFLINE badges for real-time widgets
491
- - Last update timestamps with relative time format
98
+ #### 🛠 26 Smart Tools
99
+ Consolidated from 54 tools — zero feature loss, 2× fewer IDE slots. Each tool uses an `action` parameter for multiple operations.
492
100
 
493
- ### Telegram Bot Enhancements
101
+ #### 🧠 MoE Router — 19 AI Models
102
+ Intelligent model routing with cost optimization. Supports Anthropic (Claude Opus 4.6), OpenAI (GPT-5.3 Codex), Google (Gemini 3), and Moonshot (Kimi K2.5).
494
103
 
495
- #### Added
104
+ #### 🛡️ AIDefence
105
+ <10ms threat detection: prompt injection, jailbreak, code injection, data exfiltration, social engineering. Configurable sensitivity levels.
496
106
 
497
- - **New Commands**
498
- - `/reviews` List pending code reviews with inline approve/reject buttons
499
- - `/approve [id]` — Approve a review directly from Telegram
500
- - `/reject [id] [reason]` — Reject a review with optional reason
107
+ #### 🤝 Distributed Consensus
108
+ Raft-like leader election, BFT mode, proposal system with configurable voting thresholds.
501
109
 
502
- - **Enhanced Inline Keyboards**
503
- - Reviews list shows approve/reject buttons for each pending review
504
- - Help menu now includes Reviews button
110
+ #### 🔍 HNSW Vector Search
111
+ 150×–12,500× faster than brute force. Pure TypeScript, cosine/euclidean/dot product.
505
112
 
506
- ### Auto-start Companion
113
+ #### 🌐 Cloudflare Workers
114
+ Self-hosted infrastructure: Hub, MCP Server, Telegram Bot — all on Cloudflare Free Tier.
507
115
 
508
- #### Added
116
+ #### 🔄 Full Bridge Coverage
117
+ All 26 Smart Tools work through Remote Bridge. Universal delegation via `toolName.startsWith("swarm_")`.
509
118
 
510
- - **Companion Auto-start** in `mcp-swarm-remote`
511
- - Checks if companion daemon is running on port 37373
512
- - Automatically spawns companion in background if not running
513
- - Can be disabled with `--no-companion` flag
119
+ #### 📦 One-Click Installer
120
+ `npx mcp-swarm-install` auto-detects IDEs, merges configs, supports `--telegram-user-id`.
514
121
 
515
- #### Changed
122
+ #### 🚀 Smart Router & Memory
123
+ Cost optimization (Opus → Sonnet downgrade), semantic cache, 3-tier hybrid memory system.
516
124
 
517
- - Dashboard now uses WebSocket for real-time updates instead of polling-only
518
- - Added "Real-time Monitoring" section with new widgets
519
- - Updated version to 0.9.12 in footer
125
+ #### 👥 Agent Teams & Skills
126
+ Multi-agent coordination with roles. Cross-IDE skill discovery (Gemini, Claude, Cursor, Windsurf, Codex).
520
127
 
521
128
  ---
522
129
 
523
- ## [0.9.11] - 2026-02-04
524
-
525
- ### One-Click Installer & Streamable HTTP
526
-
527
- #### Added
528
-
529
- - **One-Click Install Scripts**
530
- - `install.ps1` — PowerShell installer for Windows
531
- - `install.sh` — Bash installer for macOS/Linux
532
- - `npx mcp-swarm-install` — Interactive Node.js installer
533
-
534
- - **Streamable HTTP Transport** (MCP spec 2025-03-26)
535
- - Single `POST /mcp` endpoint instead of SSE
536
- - Session management via `Mcp-Session-Id` header
537
- - Works on Cloudflare Workers without SSE issues
538
-
539
- - **mcp-swarm-remote Proxy** (`src/remote/index.ts`)
540
- - stdio → Streamable HTTP proxy for IDE integration
541
- - Windows stdin compatibility fix
542
-
543
- #### Changed
544
-
545
- - Installer now **merges** configs instead of overwriting
546
- - Auto-detects IDE installations (Claude Desktop, Cursor, Windsurf, OpenCode, VS Code)
547
-
548
- ---
549
-
550
- ## [0.9.10] - 2026-02-03
551
-
552
- ### 🧠 MoE Router — Mixture of Experts Model Selection
553
-
554
- #### Added
555
-
556
- - **MoE Router Module** (`src/workflows/moeRouter.ts`)
557
- - Intelligent model routing based on task characteristics
558
- - Gating network for expert selection
559
- - Cost-performance optimization
560
- - Learning from feedback
561
-
562
- - **Expert Classification**
563
- - 14 task categories: code_generation, code_review, debugging, reasoning, math, creative, etc.
564
- - 4 model tiers: economy, standard, premium, flagship
565
- - 6 providers: anthropic, openai, google, mistral, local, custom
566
-
567
- - **Built-in Experts (19 models, verified pricing February 2026)**
568
- - **Anthropic Claude Series:**
569
- - Claude Opus 4.6 (flagship, 1M, $5/$25 MTok) ← NEW
570
- - Claude Opus 4.5 (flagship, 200K, $5/$25 MTok)
571
- - Claude Sonnet 4.5 (premium, 200K, $3/$15 MTok)
572
- - Claude Haiku 4.5 (economy, 200K, $1/$5 MTok)
573
- - **OpenAI GPT-5.x Series:**
574
- - GPT-5.3 Codex (flagship, 256K, ~$2/~$15 MTok) ← NEW
575
- - GPT-5.2 (flagship, 256K, $1.75/$14 MTok)
576
- - GPT-5.2 Pro (flagship, 256K, $21/$168 MTok)
577
- - GPT-5 Mini (standard, 128K, $0.25/$2 MTok)
578
- - GPT-4.1 (premium, 128K, $3/$12 MTok)
579
- - GPT-4.1 Mini (standard, 128K, $0.80/$3.20 MTok)
580
- - GPT-4.1 Nano (economy, 128K, $0.20/$0.80 MTok)
581
- - **OpenAI Reasoning Models:**
582
- - o4-mini (reasoning, 128K, $4/$16 MTok)
583
- - **Moonshot AI:**
584
- - Kimi K2.5 (premium, 128K, ~$1/~$5 MTok) ← NEW
585
- - **Google Gemini 3.x Series:**
586
- - Gemini 3 Pro (flagship, 1M, $2/$12 MTok)
587
- - Gemini 3 Flash (standard, 1M, $0.50/$3 MTok)
588
- - **Google Gemini 2.5 Series:**
589
- - Gemini 2.5 Pro (premium, 1M, $1.25/$10 MTok)
590
- - Gemini 2.5 Flash (standard, 1M, $0.30/$2.50 MTok)
591
- - Gemini 2.5 Flash Lite (economy, 1M, $0.10/$0.40 MTok)
592
-
593
- - **Routing Factors**
594
- - Task match score (category-specific strength)
595
- - Cost efficiency
596
- - Performance history (success rate)
597
- - Load balancing (prefer less recently used)
598
- - Latency constraints
599
- - Context window requirements
600
-
601
- - **Learning System**
602
- - Feedback recording
603
- - Exponential moving average for success rate
604
- - Automatic latency calibration
605
- - Per-expert statistics
606
-
607
- - **Smart Tool #54: `swarm_moe`**
608
- - `route`: Route task to best expert
609
- - `feedback`: Record routing feedback
610
- - `experts`: List available experts
611
- - `add_expert` / `remove_expert`: Manage experts
612
- - `config` / `set_config`: Configuration
613
- - `stats`: Routing statistics
614
- - `history`: Routing history
615
- - `classify`: Classify task category
616
- - `reset`: Reset statistics
617
-
618
- - **Dashboard API Endpoint** (`/api/moe`)
619
- - Routing statistics
620
- - Expert list and usage
621
- - Configuration status
622
-
623
- ---
624
-
625
- ## [0.9.9] - 2026-02-03
626
-
627
- ### 🤝 Consensus — Distributed Agreement Protocols
628
-
629
- #### Added
630
-
631
- - **Consensus Module** (`src/workflows/consensus.ts`)
632
- - Raft-like leader election with term-based leadership
633
- - Log replication for ordered command execution
634
- - Byzantine Fault Tolerance (BFT) mode for untrusted environments
635
- - Proposal system with configurable voting thresholds
636
- - Automatic failover when leader becomes unresponsive
637
-
638
- - **Consensus Modes**
639
- - `simple_majority`: 50%+ votes for approval
640
- - `raft`: Term-based leadership, log replication
641
- - `bft`: Byzantine fault tolerant (2/3+1 quorum)
642
-
643
- - **Node Management**
644
- - Join/leave cluster
645
- - Heartbeat monitoring
646
- - Trusted/untrusted node classification
647
- - Automatic dead node detection
648
-
649
- - **Proposal System**
650
- - Types: config_change, task_assignment, architecture, rollback, emergency, custom
651
- - Vote types: approve, reject, abstain
652
- - Configurable majority thresholds (0.5 to 1.0)
653
- - Expiration timeout handling
654
- - Signature verification for BFT mode
655
-
656
- - **Log Replication**
657
- - Ordered command log
658
- - Commit confirmation from leader
659
- - State machine replication across nodes
660
-
661
- - **Smart Tool #53: `swarm_consensus`**
662
- - `join`: Join consensus cluster
663
- - `leave`: Leave cluster
664
- - `heartbeat`: Send heartbeat
665
- - `status`: Get cluster status
666
- - `elect`: Start leader election
667
- - `leader`: Get current leader
668
- - `propose`: Create proposal
669
- - `vote`: Vote on proposal
670
- - `proposals`: List proposals
671
- - `get_proposal`: Get proposal details
672
- - `execute`: Execute approved proposal
673
- - `log` / `append` / `commit`: Log management
674
- - `config` / `set_config`: Configuration
675
- - `stats`: Statistics
676
-
677
- - **Dashboard API Endpoint** (`/api/consensus`)
678
- - Cluster status and node list
679
- - Leader information
680
- - Proposal statistics
681
- - Recent proposals
682
-
683
- ---
684
-
685
- ## [0.9.8] - 2026-02-03
686
-
687
- ### 🛡️ AIDefence — Security & Threat Detection
688
-
689
- #### Added
690
-
691
- - **AIDefence Module** (`src/workflows/aiDefence.ts`)
692
- - <10ms threat detection latency
693
- - Pattern-based detection with regex + heuristics
694
- - Behavioral anomaly detection
695
- - Quarantine system for suspicious content
696
- - Audit logging for security events
697
- - Configurable sensitivity levels
698
-
699
- - **Threat Categories**
700
- - `prompt_injection`: Instruction override, role hijacking, delimiter attacks
701
- - `jailbreak`: DAN mode, hypothetical bypasses, character hijacking
702
- - `code_injection`: Shell commands, eval, SQL injection, path traversal
703
- - `data_exfiltration`: API keys, credentials, external uploads
704
- - `sensitive_data`: PII, private keys, passwords
705
- - `unsafe_command`: rm -rf, sudo, crypto mining
706
- - `social_engineering`: Authority claims, urgency manipulation
707
- - `impersonation`: Agent identity spoofing
708
- - `dos_attack`: Request flooding
709
- - `unauthorized_tool`: Restricted tool access
710
-
711
- - **Sensitivity Levels**
712
- - `low`: Only critical threats
713
- - `medium`: Balanced (default)
714
- - `high`: Strict checking
715
- - `paranoid`: Maximum security
716
-
717
- - **Smart Tool #52: `swarm_defence`**
718
- - `scan`: Scan text for threats
719
- - `validate_agent`: Validate agent identity
720
- - `validate_tool`: Validate tool usage
721
- - `events`: Get security events log
722
- - `quarantine`: Get quarantined items
723
- - `release`: Release from quarantine
724
- - `stats`: Get defence statistics
725
- - `config` / `set_config`: Configuration
726
- - `trust` / `untrust`: Agent whitelist management
727
- - `clear_events`: Clear event log
728
-
729
- - **Dashboard API Endpoint** (`/api/defence`)
730
- - Security statistics
731
- - Threat distribution by category/severity
732
- - Quarantine status
733
- - Recent events log
734
-
735
- ---
736
-
737
- ## [0.9.7] - 2026-02-03
738
-
739
- ### 🔍 HNSW — Hierarchical Navigable Small World
740
-
741
- #### Added
742
-
743
- - **HNSW Module** (`src/workflows/hnsw.ts`)
744
- - Fast approximate nearest neighbor search
745
- - 150x-12,500x faster than brute force
746
- - Pure TypeScript implementation (no dependencies)
747
- - Based on Malkov & Yashunin (2016) algorithm
748
-
749
- - **Vector Operations**
750
- - Cosine similarity (default)
751
- - Euclidean distance
752
- - Dot product
753
- - Configurable dimensions (384, 768, 1536)
754
-
755
- - **Simple Embeddings**
756
- - Built-in bag-of-words embedder for demos
757
- - Works without external API
758
- - Can use custom vectors from OpenAI/Cohere/etc.
759
-
760
- - **Smart Tool #51: `swarm_vector`**
761
- - `init`: Initialize vector index
762
- - `add`: Add document with text or vector
763
- - `add_batch`: Add multiple documents
764
- - `search`: Find similar documents
765
- - `get`: Get document by ID
766
- - `delete`: Remove document
767
- - `list`: List all documents
768
- - `stats`: Index statistics
769
- - `config` / `set_config`: Configuration
770
- - `clear`: Clear entire index
771
- - `duplicates`: Find duplicate documents
772
- - `embed`: Get embedding for text
773
-
774
- - **Dashboard API Endpoint** (`/api/vector`)
775
- - Index statistics
776
- - Configuration status
777
- - Memory usage tracking
778
-
779
- #### Performance
780
-
781
- | Documents | Brute Force | HNSW | Speedup |
782
- |-----------|-------------|------|---------|
783
- | 1,000 | 10ms | 0.5ms | 20x |
784
- | 10,000 | 100ms | 0.8ms | 125x |
785
- | 100,000 | 1,000ms | 1.2ms | 833x |
786
- | 1,000,000 | 10,000ms | 2ms | 5,000x |
787
-
788
- ---
789
-
790
- ## [0.9.6] - 2026-02-03
791
-
792
- ### ⚡ Agent Booster — Fast Local Execution
793
-
794
- #### Added
795
-
796
- - **Agent Booster Module** (`src/workflows/agentBooster.ts`)
797
- - Executes trivial tasks locally without LLM API calls
798
- - 352x faster than LLM (local execution)
799
- - $0 cost (no API calls needed)
800
- - Works offline
801
- - Deterministic results
802
-
803
- - **Supported Task Types** (14 types)
804
- - `rename_variable` — Rename variables/functions across file
805
- - `fix_typo` — Fix typos in strings and comments only
806
- - `find_replace` — Simple find and replace
807
- - `add_console_log` — Add debug logging at specific line
808
- - `remove_console_log` — Remove all console.log statements
809
- - `toggle_flag` — Toggle boolean flags (true ↔ false)
810
- - `update_version` — Update version numbers
811
- - `update_import` — Update import paths
812
- - `add_comment` — Add comment at specific line
813
- - `remove_comment` — Remove single-line comments
814
- - `format_json` — Format JSON files
815
- - `sort_imports` — Sort imports alphabetically
816
- - `add_export` — Add export to function/class
817
- - `extract_constant` — Extract magic number to constant
818
-
819
- - **Smart Detection**
820
- - `can_boost` action analyzes task description
821
- - Auto-detects boostable tasks with confidence score
822
- - Extracts parameters from natural language
823
-
824
- - **Smart Tool #50: `swarm_booster`**
825
- - `execute`: Run a booster task
826
- - `can_boost`: Check if task can be boosted
827
- - `stats`: Get booster statistics
828
- - `history`: Get execution history
829
- - `config`: Get configuration
830
- - `set_config`: Update configuration
831
- - `types`: List supported task types
832
-
833
- - **Dashboard API Endpoint** (`/api/booster`)
834
- - Booster statistics and type distribution
835
- - Recent execution history
836
- - Configuration status
837
- - Cost savings tracking
838
-
839
- #### Performance
840
-
841
- | Metric | LLM | Agent Booster | Improvement |
842
- |--------|-----|---------------|-------------|
843
- | Speed | ~3000ms | ~8ms | 352x faster |
844
- | Cost | $0.01/task | $0 | 100% savings |
845
- | Offline | ❌ | ✅ | Works anywhere |
846
- | Deterministic | ⚠️ | ✅ | Same input = same output |
847
-
848
- ---
849
-
850
- ## [0.9.5] - 2026-02-03
851
-
852
- ### 🧠 SONA — Self-Optimizing Neural Architecture
853
-
854
- #### Added
855
-
856
- - **SONA Module** (`src/workflows/sona.ts`)
857
- - Self-learning task routing system inspired by Claude-Flow
858
- - Records which agents perform best for each task type
859
- - Routes new tasks to best-performing agents
860
- - Learns from outcomes with <0.05ms adaptation
861
- - Improves over time with reinforcement learning
862
-
863
- - **Task Classification**
864
- - Automatic category detection: frontend_ui, backend_api, database, testing, devops, documentation, refactoring, bugfix, feature, security, performance, infrastructure
865
- - Complexity estimation: trivial, simple, medium, complex, epic
866
- - Keyword-based pattern matching
867
- - Affected path analysis
868
-
869
- - **Agent Performance Tracking**
870
- - Success rate tracking per category
871
- - Quality score averaging
872
- - Completion time tracking
873
- - Statistical confidence calculation
874
- - Specialization detection (top 3 categories per agent)
875
-
876
- - **Online Learning**
877
- - Exponential moving average for rolling metrics
878
- - Configurable learning rate and decay
879
- - Elastic Weight Consolidation (EWC++) to prevent forgetting
880
- - Exploration/exploitation balance (10% exploration by default)
881
-
882
- - **Smart Tool #49: `swarm_sona`**
883
- - `route`: Get routing recommendation for a task
884
- - `learn`: Record task outcome and update model
885
- - `classify`: Classify a task (category, complexity)
886
- - `profile`: Get agent's performance profile
887
- - `profiles`: Get all agent profiles
888
- - `specialists`: Get top agents for a category
889
- - `history`: Get learning history
890
- - `stats`: Get SONA statistics
891
- - `config`: Get configuration
892
- - `set_config`: Update configuration
893
- - `reset`: Reset the model
894
-
895
- - **Dashboard API Endpoint** (`/api/sona`)
896
- - SONA statistics and agent profiles
897
- - Category distribution
898
- - Recent learning events
899
- - Top performing agents
900
-
901
- ---
902
-
903
- ## [0.9.4] - 2026-02-03
904
-
905
- ### 📱 Telegram Bot Integration
906
-
907
- #### Added
908
-
909
- - **Telegram Bot** (`src/integrations/telegram.ts`)
910
- - Full-featured Telegram bot for notifications and control
911
- - Event notifications: task created/completed/failed, agent joined/died, CI errors
912
- - Commands: `/status`, `/agents`, `/tasks`, `/create_task`, `/stop`, `/resume`, `/config`
913
- - Interactive inline buttons for quick actions
914
- - Priority setting via buttons (Critical, High, Medium)
915
- - Approve/Reject/Comment on reviews via Telegram
916
- - Vote on architecture decisions via Telegram
917
-
918
- - **Smart Tool #47: `swarm_telegram`**
919
- - Actions: `setup`, `config`, `enable`, `disable`, `send`
920
- - Notifications: `notify_task_created`, `notify_task_completed`, `notify_task_failed`, `notify_agent_joined`, `notify_agent_died`
921
- - Bot control: `start_polling`, `stop_polling`, `command`
922
-
923
- - **Dashboard API Endpoint** (`/api/telegram`)
924
- - Telegram configuration status
925
- - Setup instructions if not configured
926
-
927
- - **Cloudflare Worker for Telegram** (`cloudflare/telegram-bot/`)
928
- - Webhook-based (no polling required)
929
- - Deploy to Cloudflare Workers
930
- - `/setup` endpoint for webhook registration
931
-
932
- - **Smart Tool #48: `swarm_batch`**
933
- - Request batching for 50% cost savings
934
- - Supports Anthropic Message Batches API
935
- - Supports OpenAI Batch API
936
- - Actions: `queue`, `config`, `set_config`, `job`, `jobs`, `result`, `stats`, `flush`
937
-
938
- - **Batching Module** (`src/workflows/batching.ts`)
939
- - Automatic request grouping
940
- - Configurable batch size and wait time
941
- - Async batch processing with result polling
942
- - Cost savings estimation
943
-
944
- ---
945
-
946
- ## [0.9.3] - 2026-02-03
947
-
948
- ### 🧠 Smart Features & Cost Optimization
949
-
950
- #### Added
951
-
952
- - **Smart Task Routing** (`src/workflows/smartRouting.ts`)
953
- - Automatic task assignment based on file expertise
954
- - Tracks which agent edited which files
955
- - Calculates expertise scores (exact match 10x, folder 3x, extension 2x)
956
-
957
- - **Shared Context Pool** (`src/workflows/contextPool.ts`)
958
- - Agents share notes about code to avoid re-reading
959
- - Auto-staleness detection (if file hash changed)
960
- - Tags, categories, and helpful counter
961
-
962
- - **Auto Code Review** (`src/workflows/autoReview.ts`)
963
- - Automatic review assignment when task completes
964
- - Finds reviewer who knows the affected files
965
- - Comment severity levels (critical, major, minor, suggestion)
966
-
967
- - **GitHub/Linear Sync** (`src/workflows/externalSync.ts`)
968
- - Two-way sync with GitHub Issues
969
- - Linear.app integration (GraphQL API)
970
- - Auto-import issues as swarm tasks
971
- - Auto-close issues when task is done
972
-
973
- - **Cost Optimization** (`src/workflows/costOptimization.ts`)
974
- - Task complexity analysis (simple/medium/complex)
975
- - Smart model routing (cheap/standard/premium tiers)
976
- - Budget management with daily/weekly/monthly limits
977
- - Alert thresholds (50%, 80%, 95%)
978
-
979
- - **Background Heartbeat Worker** (`src/workers/`)
980
- - Uses Node.js `worker_threads` for continuous heartbeat
981
- - Works even when agent is "thinking"
982
-
983
- - **Web Dashboard** (`dashboard/`)
984
- - Real-time agent status monitoring
985
- - Orchestrator status banner with glow effects
986
- - Stats cards (agents, tasks, messages, uptime)
987
- - Task list with priority indicators
988
- - Built with Next.js + ShadCN UI
989
-
990
- - **Dashboard API** (`src/dashboardApi.ts`)
991
- - HTTP API server on port 3334
992
- - New v0.9.3 endpoints: `/api/expertise`, `/api/context`, `/api/reviews`, `/api/budget`, `/api/sync`
993
-
994
- #### Smart Tools (42-46)
995
-
996
- | # | Tool | Actions | Description |
997
- |---|------|---------|-------------|
998
- | 42 | `swarm_routing` | record, find_agent, expertise, predict, auto_assign | Smart task routing based on file expertise |
999
- | 43 | `swarm_context_pool` | add, get, search_tag, search, helpful, update, cleanup, stats | Shared context notes between agents |
1000
- | 44 | `swarm_autoreview` | create, assign, comment, complete, resolve, for_reviewer, for_author, pending, stats | Automatic code review assignment |
1001
- | 45 | `swarm_external` | enable_github, enable_linear, sync_github, sync_linear, sync_all, export_github, export_linear, status, create_issue, close_issue, comment | GitHub/Linear sync |
1002
- | 46 | `swarm_budget` | analyze, models, select, recommend, route, log_usage, usage, stats, config, set_config, check, remaining, report | Cost optimization and model routing |
1003
-
1004
- ---
1005
-
1006
- ## [0.9.1] - 2026-02-02
1007
-
1008
- ### 📚 Documentation & UX Improvements
1009
-
1010
- #### Added
1011
- - **Linux Installation Instructions** — full instructions for Linux
1012
- - **IDE-Specific Configs** — separate configurations for:
1013
- - Claude Desktop (Windows/macOS/Linux)
1014
- - Cursor, Windsurf, OpenCode CLI, VS Code + Copilot/Continue
1015
- - **Troubleshooting Section** — solutions for common issues:
1016
- - "Cannot find module" errors
1017
- - Agent not becoming orchestrator
1018
- - "repoPath is required" error
1019
- - Cloudflare Hub unavailable
1020
- - Files locked by another agent
1021
- - **Architecture Diagram** — ASCII diagram of Cloudflare Hub + Local Agents
1022
- - **Contributing Guidelines** — rules for PRs
1023
-
1024
- #### Changed
1025
- - README.md completely reworked with collapsible sections (`<details>`)
1026
-
1027
- ---
1028
-
1029
- ## [0.9.0] - 2026-02-02
1030
-
1031
- ### 🚀 MAJOR: Smart Tools Consolidation
1032
-
1033
- **Reduces 168+ individual tools → 41 Smart Tools with `action` parameter**
1034
-
1035
- #### Added
1036
-
1037
- - **Smart Tools System** — 41 unified tools replacing 168+ individual tools
1038
- - Each Smart Tool groups 3-15 related functions via `action` parameter
1039
- - Better discoverability and easier to remember
1040
- - Consistent parameter patterns across all tools
1041
-
1042
- #### Smart Tools List (41 tools)
1043
-
1044
- | # | Tool Name | Actions | Description |
1045
- |---|-----------|---------|-------------|
1046
- | 1 | `swarm_agent` | register, whoami | Agent identity |
1047
- | 2 | `swarm_task` | create, list, update, decompose, get_decomposition | Task management |
1048
- | 3 | `swarm_file` | reserve, release, list, forecast, conflicts, safety | File locking |
1049
- | 4 | `swarm_git` | sync, pr, health, cleanup, cleanup_all | Git operations |
1050
- | 5 | `swarm_worktree` | create, list, remove | Git worktrees |
1051
- | 6 | `swarm_companion` | status, stop, pause, resume | Companion daemon |
1052
- | 7 | `swarm_control` | stop, resume, status | Swarm control |
1053
- | 8 | `swarm_chat` | broadcast, dashboard, thought, thoughts | Team chat |
1054
- | 9 | `swarm_review` | request, respond, list | Code review |
1055
- | 10 | `swarm_voting` | start, vote, list, get | Voting system |
1056
- | 11 | `swarm_auction` | announce, bid, poll | Task auction |
1057
- | 12 | `swarm_mcp` | scan, authorize, policy | MCP scanning |
1058
- | 13 | `swarm_orchestrator` | elect, info, heartbeat, resign, executors, executor_heartbeat | Orchestrator |
1059
- | 14 | `swarm_message` | send, inbox, ack, reply, search, thread | Messaging |
1060
- | 15 | `swarm_briefing` | save, load | Briefings |
1061
- | 16 | `swarm_pulse` | update, get | Real-time status |
1062
- | 17 | `swarm_knowledge` | archive, search | Knowledge base |
1063
- | 18 | `swarm_snapshot` | create, rollback, list | Snapshots |
1064
- | 19 | `swarm_health` | check, dead, reassign, summary | Agent health |
1065
- | 20 | `swarm_quality` | run, report, threshold, pr_ready | Quality gate |
1066
- | 21 | `swarm_cost` | log, agent, project, limit, remaining | Cost tracking |
1067
- | 22-41 | ... | ... | Brainstorming, Plans, Debug, Spec, QA, Hooks, etc. |
1068
-
1069
- ### Backward Compatibility
1070
-
1071
- - Legacy 168+ tools server available via `npm run dev:legacy`
1072
- - Smart Tools server via `npm run dev` (default)
1073
-
1074
- ---
1075
-
1076
- ## [0.8.1] - 2026-02-02
1077
-
1078
- ### Added
1079
- - **Smart Tools Draft** — prototype of consolidating 168+ tools into 41 Smart Tools
1080
- - Files `smartTools.ts.draft` and `serverSmart.ts.draft`
1081
- - Each Smart Tool combines 3-15 similar tools via `action` parameter
1082
-
1083
- ---
1084
-
1085
- ## [0.8.0] - 2026-02-02
1086
-
1087
- ### Added
1088
- - **Orchestrator Election** (6 tools) — first agent becomes orchestrator
1089
- - `orchestrator_elect`, `orchestrator_info`, `orchestrator_heartbeat`, `orchestrator_resign`
1090
- - `executor_list`, `executor_heartbeat`
1091
-
1092
- - **Agent Messaging** (6 tools) — full inter-agent messaging system
1093
- - `agent_message_send`, `agent_inbox_fetch`, `agent_message_ack`
1094
- - `agent_message_reply`, `agent_message_search`, `agent_thread_get`
1095
-
1096
- - **Infinite Loop Mode** — orchestrator runs continuously
1097
- - Companion daemon with automatic orchestrator election
1098
- - Orchestrator does NOT stop via API — only by user
1099
- - Executors register with orchestrator
1100
- - Heartbeat system for liveness monitoring
1101
-
1102
- ---
1103
-
1104
- ## [0.7.0] - 2026-02-02
1105
-
1106
- ### Added
1107
- - **Spec Pipeline** (6 tools) — structured pipeline for specification creation
1108
- - 4 roles: gatherer → researcher → writer → critic
1109
- - **QA Loop** (7 tools) — iterative review/fix cycles
1110
- - **Guard Hooks** (6 tools) — pre-commit/pre-push safety hooks
1111
- - Bypass with `[skip-hooks]` keyword in commit message
1112
- - **Tool Clusters** (7 tools) — tool organization by categories
1113
- - 13 categories (agent, task, file, git, collab, safety, quality, debug, plan, hooks, session, cost, docs)
1114
-
1115
- ---
1116
-
1117
- ## [0.6.0] - 2026-01-30
1118
-
1119
- ### Added
1120
- - **Brainstorming Skill** (9 tools) — interactive design through step-by-step questions
1121
- - **Writing Plans Skill** (11 tools) — TDD plans with bite-sized tasks (2-5 min)
1122
- - **Systematic Debugging** (13 tools) — 4-phase debugging process
1123
- - Iron Law: NO FIXES WITHOUT ROOT CAUSE
1124
- - Red Flags detection: "Let me just try...", "Maybe if I...", "This should fix it..."
1125
-
1126
- ---
1127
-
1128
- ## [0.5.0] - 2026-01-30
1129
-
1130
- ### Added
1131
- - **Agent Health Monitor** — liveness monitoring for agents
1132
- - **Session Recording** — action recording for replay
1133
- - **Quality Gate** — automated pre-merge checks (lint, tests, types, coverage)
1134
- - **Cost Tracker** — API usage cost tracking per agent/project
1135
- - **Context Compressor** — briefing compression (ratio 0.1-0.9)
1136
- - **Regression Detector** — baseline comparison and regression detection
1137
-
1138
- ---
1139
-
1140
- ## [0.4.2] - 2026-01-28
1141
-
1142
- ### Added
1143
- - **Timeline Visualization** — task progress visualization with ASCII milestones
1144
-
1145
- ---
1146
-
1147
- ## [0.4.1] - 2026-01-25
1148
-
1149
- ### Added
1150
- - **Auto-Documentation** — automatic documentation generation on task completion
1151
- - **Agent Specialization (ML-based)** — expertise tracking per agent
1152
- - **Conflict Prediction (ML-based)** — merge conflict prediction using Git history
1153
-
1154
- ---
1155
-
1156
- ## [0.4.0] - 2026-01-20
1157
-
1158
- ### Added
1159
- - **Cloudflare Hub** — real-time WebSocket coordination
1160
- - Durable Object for state storage
1161
- - WebSocket broadcast between agents
1162
- - Hybrid mode (WS + Git fallback)
1163
- - **Orchestrator Directory** — central management (`/orchestrator/`)
1164
-
1165
- ---
1166
-
1167
- ## [0.3.0] - 2026-01-15
1168
-
1169
- ### Added
1170
- - **Collective Advice** — collective brainstorming across all agents
1171
- - **Urgent Preemption** — priority file acquisition for critical bugs
1172
- - **Snapshot & Rollback** — change rollback capability
1173
- - **Immune System** — automatic CI/test failure response
1174
-
1175
- ---
1176
-
1177
- ## [0.2.0] - 2026-01-10
1178
-
1179
- ### Added
1180
- - **Architecture Voting** — voting system for dangerous actions
1181
- - **Git Worktrees** — isolated workspaces for parallel tasks
1182
- - **GitHub Integration** — PR creation, branch sync, auto-cleanup
1183
- - **Cross-Agent Review** — code review between agents
1184
-
1185
- ---
1186
-
1187
- ## [0.1.0] - 2026-01-05
1188
-
1189
- ### Added
1190
- - **Agent Registry** — agent registration with generated names (RadiantWolf, SilentFox, etc.)
1191
- - **Task Management** — full task lifecycle (create, assign, update, complete, cancel)
1192
- - **File Locking** — exclusive/shared file locks with conflict prediction
1193
- - **Collaboration** — broadcast chat, screenshots, thought logging
1194
- - **Auction System** — task bidding for agents
1195
- - **Briefings** — mental snapshots for context transfer between agent shifts
1196
- - **Pulse** — live agent activity map
1197
- - **Knowledge Base** — findings and patterns archive
1198
- - **Ghost Mode** — autonomous code patrol and lint fixing
1199
- - **Stop Flag** — emergency stop for all agents
1200
-
1201
- ### Infrastructure
1202
- - MCP Server based on @modelcontextprotocol/sdk
1203
- - TypeScript compilation
1204
- - Installer for IDEs (Windsurf, Cursor, Claude Desktop, OpenCode, VS Code)
1205
- - Agent rules (.windsurfrules, .cursorrules, CLAUDE.md, GEMINI.md)
1206
- - Companion daemon for background tasks
1207
-
1208
- ---
1209
-
1210
- ## Semantic Versioning
1211
-
1212
- - **MAJOR (X.0.0)** — incompatible API changes
1213
- - **MINOR (0.X.0)** — new features, backward compatible
1214
- - **PATCH (0.0.X)** — bug fixes, backward compatible
1215
-
1216
- ## Legend
130
+ ### Full Changelog
1217
131
 
1218
- | Type | Description |
1219
- |------|-------------|
1220
- | **Added** | New features |
1221
- | **Changed** | Changes to existing functionality |
1222
- | **Deprecated** | Features to be removed |
1223
- | **Removed** | Removed features |
1224
- | **Fixed** | Bug fixes |
1225
- | **Security** | Vulnerability fixes |
132
+ For the complete version-by-version changelog, see the [GitHub Releases](https://github.com/AbdrAbdr/MCP-Swarm/releases).