fix: update AGENTS.md with Cognitive Assembly Line agents and pipeline commands; reset default_agent to SIN-Zeus

Author: SIN-Agent

.opencode/opencode.json

@@ -714,7 +714,7 @@
       "steps": 999999
     }
   },
-  "default_agent": "explore",
+  "default_agent": "SIN-Zeus",
   "command": {
     "sin-terminal-orchestrate": {
       "description": "Use SIN-Terminal to open and steer parallel visible OpenCode sessions",

AGENTS.md

@@ -27,11 +27,52 @@ For agent system configuration, see: https://github.com/OpenSIN-Code/Code-Swarm
 | **coder-sin-swarm** | fireworks-ai/minimax-m2.7 | Swarm Coder |
 | **Coder-SIN-Qwen** | vercel/deepseek-v4-flash | Alternative Coder |
 | **Stealth-Orchestrator** | vercel/deepseek-v4-flash | Browser Automation |
+| **explore** | stepfun-ai/step-3.5-flash | Codebase Analysis, AST-grep, Pattern Detection |
+| **orchestrator** | fireworks-ai/minimax-m2.7 | Cognitive Assembly Line Coordinator |
 
 **Simone-MCP**: All Code-Swarm agents use Simone-MCP for AST-level code operations.
 
 **If an agent you need doesn't exist: CREATE IT via `create-a2a-sin-agent` skill, no asking.**
 
+---
+
+## 🧠 Cognitive Assembly Line Pipeline
+
+The **Cognitive Assembly Line** is a 6-stage automated workflow for parallel agent execution:
+
+```
+User Prompt → Descriptor → Router → Parallel Swarms → Validation Layer → Execution Layer → Aggregation → Final Output
+```
+
+### Pipeline Commands (OpenCode.json)
+
+| Command | Purpose |
+|---------|---------|
+| `pipeline-descriptor` | Analyze prompt, generate pipeline template (complexity, subagents, time estimate) |
+| `pipeline-router` | Dispatch tasks to subagents based on descriptor |
+| `pipeline-validation` | Run code quality checks (lint, test, security, perf) |
+| `pipeline-execution` | Execute domain specialist agents |
+| `pipeline-aggregation` | Collect results, generate PRs/docs |
+| `pipeline-full` | Execute complete pipeline end-to-end |
+| `dispatch-to-subagent` | Direct dispatch to specific subagent |
+
+### Subagents (oh-my-opencode.json)
+
+The subagent layer consists of **21 specialized agents** organized in 6 groups:
+
+**Audio & Medien:** `audio_agent` (TTS/SST), `multimedia_looker` (Vision/GUI)  
+**Web-Recherche:** `athena` (Strategic), `argus` (Multi-Source), `daedalus` (Technical), `hermes_scout` (Fast Retriever)  
+**Code-Qualität:** `code-checker`, `test-runner`, `security-scanner`, `performance-auditor` (placeholders)  
+**Dokumentation:** `doc-writer`, `pr-generator`, `changelog-writer` (placeholders)  
+**DevOps:** `ci-agent`, `env-manager`, `infra-provisioner`, `backup-agent` (placeholders)  
+**Data Science:** `data-viz`, `data-analyzer`, `ml-trainer`, `ml-deployer` (placeholders)
+
+**Note:** Placeholder subagents are configured in OpenCode.json but not yet implemented. Active subagents: 13/21.
+
+See `infra-opencode-stack/README.md` for complete subagent specifications.
+
+---
+
 ## 🚀 GLOBAL BRAIN INITIALIZATION (PRIORITY -100.0)
 
 Before starting ANY work in this repo, you MUST initialize Global Brain hooks: