feat: SubAgent Swarm Topology

README.md

@@ -75,13 +75,15 @@ cp VersperClaw ~/.local/bin
 # set 200k env in ~/.claude/settings.json or terminal
 # the 20k is used to compact as buffer band
 # set tool_search env
+# set agentteam in tmux
 {
   "env": {
     "USER_TYPE": "ant",
     "CLAUDE_CODE_MAX_CONTEXT_TOKENS": "200000",
     "CLAUDE_CODE_AUTO_COMPACT_WINDOW": "200000",
     "ENABLE_TOOL_SEARCH": "true",
-    # "TAVILY_API_KEY": ""
+    "TAVILY_API_KEY": "",
+    "teammateMode": "tmux"
   }
 
   # set auto-dream config
@@ -137,16 +139,26 @@ cd desktop && bun run tauri dev
 
 ## ✨ Features
 
-### AI Friend via Omni model
+### Agent Team and View via Tmux
+
 
-<table align="center">
-  <tr>
-    <td><img src="assets/ai_friend.png"></td>
-  </tr>
-</table>
 
 ### Auto Compact & Dream & SubAgent
 
+#### 🧠 Core Mechanisms of Cycle Long Run
+
+* **Auto Compact** — Automatically optimizes and compresses historical context to bypass sequence length bottlenecks.
+* **Dream** — Consolidates raw operational data into structured, long-term parametric insights asynchronously.
+
+#### 🗂️ SubAgent Swarm Topology
+
+| SubAgent Type | Core Responsibility | Operational Gate |
+| :--- | :--- | :--- |
+| 🛡️ **General-Purpose** | Orchestrates routing, high-level user interaction, and dispatching. | Active |
+| 🔍 **Explore** | Conducts deep semantic search and autonomous tool discovery. | Active |
+| 📋 **Plan** | Handles multi-step chain-of-thought strategy and task decomposition. | Active |
+| 🧪 **Verification** | Executes automated runtime validation, assertion checks, and fault tolerance. | Active |
+
 <table align="center">
   <tr>
     <td><img src="assets/ctx_auto_compose/compact_dream_subagent.png"></td>
@@ -163,3 +175,10 @@ cd desktop && bun run tauri dev
   </tr>
 </table>
 
+### AI Friend via Omni model
+
+<table align="center">
+  <tr>
+    <td><img src="assets/ai_friend.png"></td>
+  </tr>
+</table>

src/tools/AgentTool/builtInAgents.ts

@@ -11,6 +11,10 @@ import { VERIFICATION_AGENT } from './built-in/verificationAgent.js'
 import type { AgentDefinition } from './loadAgentsDir.js'
 
 export function areExplorePlanAgentsEnabled(): boolean {
+  if (process.env.USER_TYPE !== 'ant') {
+    return true
+  }
+
   if (feature('BUILTIN_EXPLORE_PLAN_AGENTS')) {
     // 3P default: true — Bedrock/Vertex keep agents enabled (matches pre-experiment
     // external behavior). A/B test treatment sets false to measure impact of removal.
@@ -19,6 +23,17 @@ export function areExplorePlanAgentsEnabled(): boolean {
   return false
 }
 
+function isVerificationAgentEnabled(): boolean {
+  if (process.env.USER_TYPE !== 'ant') {
+    return true
+  }
+
+  if (feature('VERIFICATION_AGENT')) {
+    return getFeatureValue_CACHED_MAY_BE_STALE('tengu_hive_evidence', false)
+  }
+  return false
+}
+
 export function getBuiltInAgents(): AgentDefinition[] {
   // Allow disabling all built-in agents via env var (useful for SDK users who want a blank slate)
   // Only applies in noninteractive mode (SDK/API usage)
@@ -61,10 +76,7 @@ export function getBuiltInAgents(): AgentDefinition[] {
     agents.push(VERSPER_CLAW_GUIDE_AGENT)
   }
 
-  if (
-    feature('VERIFICATION_AGENT') &&
-    getFeatureValue_CACHED_MAY_BE_STALE('tengu_hive_evidence', false)
-  ) {
+  if (isVerificationAgentEnabled()) {
     agents.push(VERIFICATION_AGENT)
   }