Update subagent model selection guidance

- Use kimi-k2.5 for opus-level tasks
- Use nemotron-3-super for sonnet-level tasks
- Use kimi-k2-thinking for deep reasoning at flash end
- Document what NOT to use (qwen, claude,)

Signed: Annie Tunturi <ani@wiuf.net>

👾 Generated with [Letta Code](https://letta.com)

Co-Authored-By: Letta Code <noreply@letta.com>

system/subagent_usage_guide.md

@@ -26,10 +26,24 @@ When running subagents via the Task tool, always specify a valid model from the
 - `nemotron-3-super` - NVIDIA model
 - `gpt-oss-120b` - OpenAI OSS
 
-### Preferred Models
-- **For exploration/research**: `kimi-k2.5` or `kimi-k2.5-nvfp4`
-- **For coding tasks**: `qwen3-coder` or `glm-4.7-flash`
-- **For complex reasoning**: `kimi-k2-thinking` or `deepseek-r1`
+### Tiered Model Selection
+
+| Task Level | Model | Purpose |
+|------------|-------|---------|
+| **Opus-level** | `kimi-k2.5` | Deep research, complex analysis, high-stakes decisions |
+| **Sonnet-level** | `nemotron-3-super` | Mid-complexity, balanced reasoning, implementation |
+| **Deep reasoning** | `kimi-k2-thinking` | Verification, synthesis, final review at end of chain |
+
+### What NOT to Use
+- **NO `qwen`** - User preference: "Qwen sucks"
+- **NO `claude` models** - Hardcoded to proxies, avoid confusion
+- **NO `glm-4.7-flash`** - Use `nemotron-3-super` instead for lightweight tasks
+
+### Preferred Models (Updated)
+- **Opus-level tasks**: `kimi-k2.5` (heavy research, architecture decisions)
+- **Sonnet-level tasks**: `nemotron-3-super` (implementation, exploration)
+- **Deep reasoning**: `kimi-k2-thinking` (verification, synthesis)
+- **Fast exploration**: `kimi-k2.5-nvfp4` (quick searches)
 
 ## Common Errors
 
@@ -89,3 +103,4 @@ Task({
 
 Last updated: 2026-03-21
 Discovered during: Matrix bridge formatting investigation
+Signed: Annie Tunturi <ani@wiuf.net>