fix: disable desktop context injection by default (#95) (#97)

* fix: disable desktop context injection by default (#95)

The proxy was injecting a ~1500 token desktop context system prompt
into every request, inflating prompt_tokens from ~17 to ~1587 for
a simple message. Add `model.inject_desktop_context` config option
(default false) to skip injection unless explicitly enabled.

* docs: update CHANGELOG for #95

* chore: revert unrelated default.yaml changes

---------

Co-authored-by: icebear0828 <icebear0828@users.noreply.github.com>

CHANGELOG.md

@@ -6,6 +6,10 @@
 
 ## [Unreleased]
 
+### Fixed
+
+- 默认关闭 desktop context 注入：之前每次请求注入 ~1500 token 的 Codex Desktop 系统提示，导致 prompt_tokens 虚高；新增 `model.inject_desktop_context` 配置项（默认 `false`），需要时可手动开启 (#95)
+
 ### Added
 
 - 额度自动刷新 + 分层预警：后台每 5 分钟（可配置）定时拉取所有账号的官方额度，缓存到 AccountEntry 供 Dashboard 即时读取；额度达到阈值（默认 80%/90%，可自定义）时显示 warning/critical 横幅；额度耗尽的账号自动标记为 rate_limited 跳过分配，到期自动恢复 (#92)

config/default.yaml

@@ -12,6 +12,7 @@ model:
   default: gpt-5.2-codex
   default_reasoning_effort: medium
   default_service_tier: null
+  inject_desktop_context: false
   suppress_desktop_directives: true
 auth:
   jwt_token: null

src/config.ts

@@ -23,6 +23,7 @@ const ConfigSchema = z.object({
     default: z.string().default("gpt-5.2-codex"),
     default_reasoning_effort: z.string().default("medium"),
     default_service_tier: z.string().nullable().default(null),
+    inject_desktop_context: z.boolean().default(false),
     suppress_desktop_directives: z.boolean().default(true),
   }),
   auth: z.object({

src/translation/shared-utils.ts

@@ -43,6 +43,7 @@ const SUPPRESS_PROMPT =
  * to override desktop-specific behaviors.
  */
 export function buildInstructions(userInstructions: string): string {
+  if (!getConfig().model.inject_desktop_context) return userInstructions;
   const ctx = getDesktopContext();
   if (!ctx) return userInstructions;
   if (getConfig().model.suppress_desktop_directives) {