Sync public snapshot from freebuff-private

Source: CodebuffAI/freebuff-private@fa4b28d7fe2df4088cb31f95fe9ba5b4aba52d4c

Author: github-actions[bot]

agents/thinker/thinker-gemini.ts

@@ -1,4 +1,4 @@
-import { FREEBUFF_DEEPSEEK_V4_FLASH_MODEL_ID } from '@codebuff/common/constants/freebuff-models'
+import { FREEBUFF_GPT_5_6_LUNA_MODEL_ID } from '@codebuff/common/constants/freebuff-models'
 
 import thinker from './thinker'
 
@@ -7,30 +7,21 @@ import type { SecretAgentDefinition } from '../types/secret-agent-definition'
 /**
  * The freebuff.com/chat thinker child, spawned by base-chat.
  *
- * MOVED OFF GEMINI 3.1 PRO ON 2026-09-01. The Google credit grant that paid for
- * it ran out, and at list price this one agent was ~$2,700/week — nine times
- * the chat root models combined — because every spawn re-reads the whole
- * conversation (~52k input tokens a call) at $2/M. DeepSeek V4 Flash at the
- * same token volume is ~$70-85/week, runs on our own DeepSeek direct and
- * Luminal lanes rather than Merge, and thinks natively.
+ * Uses GPT-5.6 Luna with extra-high reasoning for Chat deliberation.
  *
  * The id stays `thinker-gemini`: chat/agent.ts registers it by import, the
  * hidden-agent list and FREEBUFF_GEMINI_PRO_AGENT_IDS name it, and base-chat's
  * prompt spawns it by name. Renaming buys nothing and touches all of them.
- *
- * Reasoning is declared explicitly rather than left to the catalog default
- * (also `high`): the DeepSeek lane maps `reasoning` onto its `thinking` flag,
- * and a thinker that does not think is just a slower copy of the root.
  */
 const definition: SecretAgentDefinition = {
   ...thinker,
   id: 'thinker-gemini',
   displayName: 'Thinker',
-  model: FREEBUFF_DEEPSEEK_V4_FLASH_MODEL_ID,
+  model: FREEBUFF_GPT_5_6_LUNA_MODEL_ID,
   providerOptions: undefined,
   reasoningOptions: {
     enabled: true,
-    effort: 'high',
+    effort: 'xhigh',
   },
   outputSchema: undefined,
   outputMode: 'last_message',

agents/types/agent-definition.ts

@@ -47,7 +47,7 @@ export interface AgentDefinition {
         max_tokens: number
       }
     | {
-        effort: 'high' | 'medium' | 'low' | 'minimal' | 'none'
+        effort: 'max' | 'xhigh' | 'high' | 'medium' | 'low' | 'minimal' | 'none'
       }
   )