Merge upstream main into fix/sdk-list-directory-path-boundary

Author: c8dhjp4tyv-bit

.gitignore

@@ -2,6 +2,9 @@
 .idea/
 .vercel
 
+# Exported user lists (PII) - scripts/export-user-emails.ts
+user-emails-*.csv
+
 # Environment files (secrets) - Bun loads .env.* files natively
 .env
 .env.*
@@ -25,6 +28,9 @@ npm-app/src/__tests__/data/
 debug/
 .context/
 docs/bot-detection.md
+# Issue-specific plans/specs belong in Linear, not the repository. Existing
+# tracked historical specs remain tracked; this keeps new local artifacts out.
+docs/specs/
 .codex
 
 # Nx cache directories
@@ -66,3 +72,13 @@ scratch-*
 
 # Ad-hoc investigation scripts — scratch, never committed
 scripts/_tmp-*
+
+# Local operator launch manifests contain real advertiser identifiers. The
+# checked-in example is intentionally non-routable.
+ads-pilot-launch.local.json
+
+# Linked git worktrees created by agent tooling. Each is a full checkout with
+# its own .git file, so committing one nests a second working tree inside this
+# one — and because a worktree carries its own generated migrations, it also
+# drags a DUPLICATE migration index into whichever branch swept it up.
+.claude/worktrees/

README.md

@@ -34,12 +34,15 @@ Freebuff includes a curated model catalog. The regular picker currently offers:
 
 | Model                       | Access                  | Best for                                                          |
 | --------------------------- | ----------------------- | ----------------------------------------------------------------- |
-| **DeepSeek V4 Flash 07/31** | Full access             | The default everywhere in full mode; fast coding and tool use     |
-| **GPT-5.6 Luna**            | Full access             | Strong all-around with native images; two sessions a day          |
+| **GLM 5.3 Flash**           | Full access             | The default in full mode; deepest reasoning, unmetered            |
+| **GPT-5.6 Luna**            | Full access             | Strong all-around with native images                              |
+| **DeepSeek V4 Flash 07/31** | Full access             | Fast coding and tool use; pauses during peak hours                |
 | **MiMo 2.5**                | Full and limited access | The limited-mode default; balanced performance with image support |
-| **DeepSeek V4 Pro**         | Full access             | Deepest reasoning; one session a day                              |
+| **Solar Pro 4**             | Full access             | Limited-time trial; 524K context, text only                       |
 
-These limits are **temporary**, and they exist because the providers serving DeepSeek now charge more than free mode can carry. V4 Pro is one session a day; GPT-5.6 Luna is two sessions a day; models may serve from a quantized (Q8_0) build. MiMo 2.5 stays unlimited. All of it is intended to be reverted.
+Most models draw on your normal daily sessions rather than a separate limit. GLM 5.3 Flash and MiMo 2.5 are unmetered and cost no session at all. Models may still serve from a quantized (Q8_0) build.
+
+DeepSeek V4 Pro was retired from the catalog; GLM 5.3 Flash replaces it as the deep-reasoning pick.
 
 Beyond the regular picker:
 

README.zh-CN.md

@@ -34,12 +34,15 @@ Freebuff 提供经过筛选的模型目录。常规模型选择器目前包括
 
 | 模型                        | 访问范围       | 适用场景                                         |
 | --------------------------- | -------------- | ------------------------------------------------ |
-| **DeepSeek V4 Flash 07/31** | 完整访问       | 完整模式下所有平台的默认模型；快速编程和工具调用 |
-| **GPT-5.6 Luna**            | 完整访问       | 综合能力强，原生支持图像；每天两次会话           |
+| **GLM 5.3 Flash**           | 完整访问       | 完整模式下的默认模型；推理最深入，且不消耗会话   |
+| **GPT-5.6 Luna**            | 完整访问       | 综合能力强，原生支持图像                         |
+| **DeepSeek V4 Flash 07/31** | 完整访问       | 快速编程和工具调用；高峰时段暂停                 |
 | **MiMo 2.5**                | 完整和受限访问 | 受限模式的默认模型；均衡性能并支持图像           |
-| **DeepSeek V4 Pro**         | 完整访问       | 推理最深入；每天一次会话                         |
+| **Solar Pro 4**             | 完整访问       | 限时试用；52.4 万上下文，仅支持文本             |
 
-以下限制是**临时**的，原因是为 DeepSeek 提供服务的供应商收费已超出免费模式可承受的范围。V4 Pro 每天一次会话；GPT-5.6 Luna 每天两次会话；模型可能由量化（Q8_0）版本提供服务。MiMo 2.5 保持无限使用。这些调整都计划在之后恢复。
+大多数模型使用你的常规每日会话，而不再各自设限。GLM 5.3 Flash 和 MiMo 2.5 保持无限使用，完全不消耗会话。模型仍可能由量化（Q8_0）版本提供服务。
+
+DeepSeek V4 Pro 已从模型目录中下线，由 GLM 5.3 Flash 接替其深度推理的位置。
 
 常规模型选择器之外：
 

agents/__tests__/base3.test.ts

@@ -12,9 +12,12 @@ import base3FreeDeepseekFlash from '../base3-free-deepseek-flash'
 import base3FreeDeepseekFlashEvals from '../base3-free-deepseek-flash-evals'
 import base3FreeFable from '../base3-free-fable'
 import base3FreeGlm from '../base3-free-glm'
+import base3FreeGlmV53Flash from '../base3-free-glm-5-3-flash'
 import base3FreeLuna from '../base3-free-luna'
 import base3FreeMimo from '../base3-free-mimo'
 import base3FreeMinimaxM3 from '../base3-free-minimax-m3'
+import base3FreeOxAlpha from '../base3-free-ox-alpha'
+import base3FreeSolarPro4 from '../base3-free-solar-pro4'
 import base3Lite from '../base3-lite'
 
 /**
@@ -44,13 +47,18 @@ const CLI_ROOTS = [
   base3FreeMinimaxM3,
   base3FreeMimo,
   base3FreeGlm,
+  base3FreeGlmV53Flash,
   base3FreeLuna,
   base3FreeFable,
+  base3FreeOxAlpha,
+  base3FreeSolarPro4,
 ]
 
 describe('base3 CLI roots', () => {
   test('keeps the efficiency flags the runtime reads', () => {
-    expect(CLI_ROOTS.length).toBe(11)
+    // 14 since Solar Pro 4 reached the CLI. The count is asserted so a root
+    // added without the flags below cannot slip in unnoticed.
+    expect(CLI_ROOTS.length).toBe(14)
     for (const agent of CLI_ROOTS) {
       // Windowed reads + the 100-entry glob cap + search-first tool wording.
       expect(agent.windowedFileReads).toBe(true)
@@ -83,6 +91,9 @@ describe('base3 CLI roots', () => {
     // requires a canonical opening at byte 0, so prepending 403s every turn.
     for (const agent of CLI_ROOTS) {
       expect(hasFreebuffRootSystemPromptOpening(agent.systemPrompt!)).toBe(true)
+      expect(
+        agent.systemPrompt!.match(/\{CODEBUFF_GIT_CHANGES_PROMPT\}/g),
+      ).toHaveLength(1)
     }
   })
 

agents/base-chat.ts

@@ -66,6 +66,10 @@ End every response by calling the suggest_followups tool with exactly 3 followup
       'meta/muse-spark-1.2-contributor': 1_000_000,
       // Ox Alpha: 1,048,576 published, entered low for the same reason.
       'stealth/ox-alpha': 1_000_000,
+      // GLM 5.3 Flash: 1,310,720 published, entered low for the same reason.
+      'z-ai/glm-5.3-flash': 1_000_000,
+      // Solar Pro 4: 524,288 published, entered low for the same reason.
+      'upstage/solar-pro4': 500_000,
     }
 
     /** For any model not listed above. Assuming a window is smaller than it is

agents/base2/base2-free-glm-5-3-flash.ts

@@ -0,0 +1,13 @@
+import { FREEBUFF_GLM_V53_FLASH_MODEL_ID } from '@codebuff/common/constants/freebuff-models'
+
+import { createBase2 } from './base2'
+
+const definition = {
+  ...createBase2('free', {
+    model: FREEBUFF_GLM_V53_FLASH_MODEL_ID,
+  }),
+  id: 'base2-free-glm-5-3-flash',
+  displayName: 'Buffy the GLM 5.3 Flash Free Orchestrator',
+}
+
+export default definition

agents/base2/base2-free-solar-pro4.ts

@@ -0,0 +1,13 @@
+import { FREEBUFF_SOLAR_PRO_4_MODEL_ID } from '@codebuff/common/constants/freebuff-models'
+
+import { createBase2 } from './base2'
+
+const definition = {
+  ...createBase2('free', {
+    model: FREEBUFF_SOLAR_PRO_4_MODEL_ID,
+  }),
+  id: 'base2-free-solar-pro4',
+  displayName: 'Buffy the Solar Pro 4 Free Orchestrator',
+}
+
+export default definition

agents/base3-free-glm-5-3-flash.ts

@@ -0,0 +1,19 @@
+import { FREEBUFF_GLM_V53_FLASH_MODEL_ID } from '@codebuff/common/constants/freebuff-models'
+
+import { createBase3CliRoot } from './base3'
+
+// No `reasoningOptions`, for the reason spelled out on base3-free-luna: an
+// agent-declared reasoning reaches the wire as `body.reasoning`, which makes
+// the definition rather than the user the authority on effort. It costs nothing
+// here in particular — OpenRouter reports no effort levels for this model, so
+// the catalog row declares no ladder and the picker shows no control.
+const definition = {
+  ...createBase3CliRoot({
+    model: FREEBUFF_GLM_V53_FLASH_MODEL_ID,
+    isFreebuff: true,
+  }),
+  id: 'base3-free-glm-5-3-flash',
+  displayName: 'Buffy on GLM 5.3 Flash',
+}
+
+export default definition

agents/base3-free-ox-alpha.ts

@@ -0,0 +1,28 @@
+import { FREEBUFF_OX_ALPHA_MODEL_ID } from '@codebuff/common/constants/freebuff-models'
+
+import { createBase3CliRoot } from './base3'
+
+/**
+ * Buffy on Ox Alpha, for CLI and Desktop (2026-08-24).
+ *
+ * No `reasoningOptions`, for the reason spelled out on base3-free-luna: an
+ * agent-declared reasoning reaches the wire as `body.reasoning` and makes the
+ * agent the authority on effort, which leaves the user's picker choice unable
+ * to do anything. That matters more here than on most rows -- this endpoint
+ * reports `reasoning.mandatory: true` with `default_effort: max`, and `max`
+ * costs ~4.8x the tokens of `high` for no better answer while being the rung
+ * most likely to spend the whole budget thinking and return
+ * `finish_reason: "length"` with null content. The catalog row names `high`
+ * explicitly and the server fills it in; hard-coding anything here would take
+ * that away.
+ */
+const definition = {
+  ...createBase3CliRoot({
+    model: FREEBUFF_OX_ALPHA_MODEL_ID,
+    isFreebuff: true,
+  }),
+  id: 'base3-free-ox-alpha',
+  displayName: 'Buffy on Ox Alpha',
+}
+
+export default definition

agents/base3-free-solar-pro4.ts

@@ -0,0 +1,14 @@
+import { FREEBUFF_SOLAR_PRO_4_MODEL_ID } from '@codebuff/common/constants/freebuff-models'
+
+import { createBase3CliRoot } from './base3'
+
+const definition = {
+  ...createBase3CliRoot({
+    model: FREEBUFF_SOLAR_PRO_4_MODEL_ID,
+    isFreebuff: true,
+  }),
+  id: 'base3-free-solar-pro4',
+  displayName: 'Buffy on Solar Pro 4',
+}
+
+export default definition