feat: evolutionary agent model upgrades based on recommendation matrix

- devops-engineer: deepseek-v3.2 → kimi-k2.6:cloud (★88)
- browser-automation: glm-5 → kimi-k2.6:cloud (★86)
- visual-tester: glm-5 → qwen3-coder:480b (★82)
- agent-architect: nemotron-3-super → kimi-k2.6:cloud (★86)
- orchestrator: glm-5 → kimi-k2.6:cloud (dispatch critical)
- product-owner: glm-5 → glm-5.1 (★84)
- prompt-optimizer: qwen3.6-plus:free → glm-5.1 (stable fallback)
- system-analyst: qwen3.6-plus:free → glm-5.1 (★90)
- Add autonomous-mode.md rule for zero-confirmation workflow

kilo.jsonc

@@ -313,7 +313,7 @@
     "prompt-optimizer": {
       "description": "Improves agent system prompts based on performance failures. Meta-learner for prompt optimization",
       "mode": "subagent",
-      "model": "qwen/qwen3.6-plus:free",
+      "model": "ollama-cloud/glm-5.1",
       "permission": {
         "read": "allow",
         "edit": "allow",