docs(AGENT-SETUP): Add real-world configuration examples with SDD pipeline and GSR

[RamonsDka]

Pre-flight Checks

• I have searched existing issues and this is not a duplicate
• I have a link to the specific doc page or section that needs improvement

Which Document?

docs/AGENT-SETUP.md

What is Wrong or Missing?

The current docs show basic per-agent configurations (Claude Code, OpenCode, Gemini CLI, Codex, VS Code, Cursor, Windsurf). These are great for simple setups. However, there are no examples of real-world complex configurations that power users actually build.

What's missing:

1. Model routing integration: Users who use model routers (like GSR — Gentle SDD Router) that dynamically assign different models to different agents need to know how engram sessions track which model was active.

2. Sub-agent pipeline: OpenCode supports sub-agents (mode: subagent) that are invoked via delegation. Each sub-agent might have:

   • Its own model
   • Its own fallback chain
   • Shared engram memory with the parent

   The docs don't show how to configure this.

3. Plugin configuration: The PLUGINS.md mentions OpenCode plugins but doesn't show the actual opencode.json MCP configuration for complex setups with multiple providers.

4. Profile-based configurations: Users who maintain multiple profiles (e.g., "free tier", "mixed", "premium") need engram to work consistently across profile switches. No docs cover this.

Suggested Improvement

Add a "Real-World Configuration Examples" section to docs/AGENT-SETUP.md with complete, copy-pasteable examples:

Example 1: OpenCode with SDD Pipeline

{
  "mcp": {
    "engram": {
      "command": ["engram", "mcp", "--project=my-project"],
      "type": "local"
    }
  }
}

With the corresponding AGENTS.md protocol section for the orchestrator and sub-agents.

Example 2: OpenCode with GSR Model Router

{
  "mcp": {
    "engram": {
      "command": ["engram", "mcp", "--project=my-project"],
      "type": "local"
    }
  },
  "agent": {
    "gsr-mixto": {
      "mode": "primary",
      "model": "openai/gpt-5.3-instant",
      "fallback": "mistral-large-3, glm-5",
      "prompt": "{file:./AGENTS.md}"
    }
  }
}

Show how the engram protocol in AGENTS.md should instruct the agent to use engram regardless of which model is active.

Example 3: Multi-project with session isolation

Show how --project flag creates isolated memory spaces and how to configure the MCP command per-project.

Example 4: Claude Code + OpenCode sharing engram

Both agents on the same machine sharing the same engram instance. Show the config for both.

Additional Context

Working examples of all of these exist in: https://github.com/RamonsDka/MI-FULL-CONFIG-OPENCODE-TOOLS

The repo has complete configs for:

• 10+ SDD sub-agents with individual models and fallbacks
• 3 GSR presets (free/mixed/premium) with model routing
• 8 TUI themes
• Profile management system
• All sharing a single engram instance

Would be happy to contribute these as documentation PRs if approved.

[QADovah]

Hi 5 to this. Just found out, reading issues, when I check my engram stats, I only have 1 session within only 1/6 projects. Find really hard to try to understand how to set the clockwork on motion.