feat(council): define litellm-backed tier policy

MEMORY.md

@@ -33,6 +33,7 @@
 - Use local-first search by default: SearXNG first, then Brave-backed fallback when needed.
 - Brave free-plan search is rate-limited heavily; avoid parallel bursts.
 - In direct sessions with Will, cron jobs should use the `automation` agent by default unless Will explicitly says otherwise in that session.
+- Council tiers should use local LiteLLM-backed models for usage monitoring: light = `litellm/gpt-5.3-codex` with low thinking, medium = `litellm/gpt-5.3-codex` with high thinking, heavy = `litellm/gpt-5.4` with high thinking.
 
 ## Infrastructure notes worth remembering
 - Full `~/.openclaw` backups upload to MinIO bucket `zap` and are scheduled via OS cron every 6 hours.

skills/council/SKILL.md

@@ -204,7 +204,13 @@ Before orchestrating a council run, resolve the plan first, for example:
 - `python3 skills/council/scripts/council-plan.py --mode personality --tier light --pretty`
 - `python3 skills/council/scripts/council-plan.py --mode dp --tier heavy --pretty`
 
-Then use the returned `agentId` values in `sessions_spawn`. If `modelOverride` is null, preserve the dedicated agent's configured model. If present, apply the override only for that run.
+Then use the returned `agentId` values in `sessions_spawn`.
+- Apply `modelOverride` for that run.
+- Apply `thinkingOverride` for that run.
+- Current council policy uses local LiteLLM-backed models for monitoring/usage visibility:
+  - `light`  -> `litellm/gpt-5.3-codex` with `thinking=low`
+  - `medium` -> `litellm/gpt-5.3-codex` with `thinking=high`
+  - `heavy`  -> `litellm/gpt-5.4` with `thinking=high`
 
 ## Configuration
 

skills/council/references/roster.json

@@ -56,19 +56,25 @@
   },
   "tierPolicy": {
     "light": {
-      "advisorModel": null,
+      "advisorModel": "litellm/gpt-5.3-codex",
-      "synthesisModel": null,
+      "synthesisModel": "litellm/gpt-5.3-codex",
-      "notes": "Use dedicated council agents as configured. No overrides by default."
+      "advisorThinking": "low",
+      "synthesisThinking": "low",
+      "notes": "Use local LiteLLM for all council roles. Light tier = gpt-5.3-codex with low thinking."
     },
     "medium": {
-      "advisorModel": null,
+      "advisorModel": "litellm/gpt-5.3-codex",
-      "synthesisModel": null,
+      "synthesisModel": "litellm/gpt-5.3-codex",
-      "notes": "Keep advisor agents cheap; rely on stronger dedicated synthesis agents."
+      "advisorThinking": "high",
+      "synthesisThinking": "high",
+      "notes": "Use local LiteLLM for all council roles. Medium tier = gpt-5.3-codex with high thinking."
     },
     "heavy": {
-      "advisorModel": null,
+      "advisorModel": "litellm/gpt-5.4",
-      "synthesisModel": "openai-codex/gpt-5.4",
+      "synthesisModel": "litellm/gpt-5.4",
-      "notes": "Preserve dedicated role identities. Escalate synthesis first; only override advisors when task risk justifies it."
+      "advisorThinking": "high",
+      "synthesisThinking": "high",
+      "notes": "Use local LiteLLM for all council roles. Heavy tier = gpt-5.4 with high thinking."
     }
   }
 }

skills/council/scripts/council-plan.py

@@ -29,12 +29,15 @@ def resolve_plan(mode: str, tier: str):
 
     for role, cfg in roles.items():
         mission = cfg.get('mission', 'advisor')
-        override = tier_policy.get('synthesisModel') if mission == 'synthesis' else tier_policy.get('advisorModel')
+        is_synthesis = mission == 'synthesis'
+        override = tier_policy.get('synthesisModel') if is_synthesis else tier_policy.get('advisorModel')
+        thinking = tier_policy.get('synthesisThinking') if is_synthesis else tier_policy.get('advisorThinking')
         plan['roles'][role] = {
             'agentId': cfg['agentId'],
             'mission': mission,
             'defaultModel': cfg.get('defaultModel'),
             'modelOverride': override,
+            'thinkingOverride': thinking,
             'fallbacks': cfg.get('fallbacks', [])
         }
     return plan