GitLab

对齐/v1/responses的Forward方法，修复两处不一致：
- 成功响应时从响应头提取OAuth账号的Codex使用量数据
- 非failover错误场景下应用管理员配置的错误透传规则

- Fix gofmt: align struct field comments in UsageCache, trim trailing
  whitespace on const comments
- Fix errcheck: use comma-ok on type assertion for singleflight result
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Prevents 429 rate-limit retry storms and reduces upstream correlation risk
for Anthropic usage API queries.

Three changes:
1. Negative caching (1 min TTL) — 429/error responses are now cached,
   preventing every subsequent page load from re-triggering failed API calls.
2. singleflight dedup — concurrent requests for the same account are
   collapsed into a single upstream call, preventing cache stampede.
3. Random jitter (0–800 ms) — staggers multi-account cache-miss bursts so
   requests from different accounts don't hit upstream simultaneously with
   identical TLS fingerprints, reducing anti-abuse correlation risk.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

从响应头 anthropic-ratelimit-unified-5h-utilization 获取并存储真实
utilization 值，解决进度条始终显示 0% 的问题。窗口重置时清除旧值，
避免残留上个窗口的数据。
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

移除订阅扣费 stub 中未被使用的状态字段与赋值，
消除 golangci-lint 的 unused 告警，保持回归测试语义不变。
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

抽取共享的用户分组专属倍率解析器，统一缓存、singleflight 与回退逻辑。\n\n让 OpenAI 独立计费链路复用专属倍率解析，修复 usage 记录与实际扣费未命中用户专属倍率的问题。\n\n补齐 OpenAI 计费与解析器单元测试，并修复全量回归中暴露的 lint 阻塞项。\n\nCo-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

commit 80ae592c 引入 lite 模式优化首次加载性能，但将窗口费用查询也一起跳过了。
commit 491a7444

 尝试用 30 秒快照缓存修复，但缓存过期后问题复现。

移除窗口费用查询的 lite/非 lite 区分，始终执行 PostgreSQL 聚合查询。
同时删除不再需要的 account_window_cost_cache.go 文件。
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

- Add recoverAnthropicMessagesPanic for Messages handler to return
  Anthropic-formatted errors instead of OpenAI Responses format on panic
- Add nil check for rateLimitService.HandleUpstreamError in
  ForwardAsAnthropic to match defensive pattern used elsewhere
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Add Anthropic Messages API support for OpenAI platform groups, enabling
clients using Claude-style /v1/messages format to access OpenAI accounts
through automatic protocol conversion.

- Add apicompat package with type definitions and bidirectional converters
  (Anthropic ↔ Chat, Chat ↔ Responses, Anthropic ↔

 Responses)
- Implement /v1/messages endpoint for OpenAI gateway with streaming support
- Add model mapping UI for OpenAI OAuth accounts (whitelist + mapping modes)
- Support prompt caching fields and codex OAuth transforms
- Fix tool call ID conversion for Responses API (fc_ prefix)
- Ensure function_call_output has non-empty output field
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

透传模式账号仅替换认证，应允许所有模型通过。之前调度阶段的
isModelSupportedByAccount 不感知透传模式，导致 model_mapping
中未配置的新模型（如 gpt-5.4）被拒绝返回 503。

lite 模式下从快照缓存读取窗口费用，非 lite 模式查询后写入缓存

补齐 gpt-5.4 fallback 的长上下文计费元信息，\n确保超过 272000 输入 token 时对整次会话应用\n2x 输入与 1.5x 输出计费规则。\n\n同时将官方快照 gpt-5.4-2026-03-05 加入前端\n白名单候选与回归测试，避免 whitelist 模式误拦截。\n\nCo-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

(cherry picked from commit d95497af87f608c6dadcbe7d6e851de9413ae147)

- 接入 gpt-5.4 模型识别与规范化，补充默认模型列表
- 增加 gpt-5.4 输入/缓存命中/输出价格与计费兜底逻辑
- 同步前端模型白名单与 OpenCode 上下文窗口（1050000/128000）
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
(cherry picked from commit 924476dcac6181cd0f3ee731ec7b73672ff03793)

LoadFactor should only affect scheduling weight, not load rate reporting.

- Fix bulk edit: send 0 instead of null/NaN to clear load_factor
- Fix edit modal: explicit NaN check instead of implicit falsy
- Fix create modal: use ?? instead of || for load_factor
- Add load_factor upper limit validation (max 10000)
- Add //go:build unit tag and self-contained intPtrHelper in test
- Add design intent comments on WaitPlan.MaxConcurrency

The test could flake depending on exact execution time near midnight
boundaries. Use a range check (29 or 30) instead of exact equality.

- Extract postUsageBilling() to consolidate billing logic across
  GatewayService.RecordUsage, RecordUsageWithLongContext, and
  OpenAIGatewayService.RecordUsage, eliminating ~120 lines of
  duplicated code
- Fix account quota to use TotalCost × accountRateMultiplier
  (was using raw TotalCost, inconsistent with account cost stats)
- Fix RecordUsageWithLongContext API Key quota only updating in
  balance mode (now updates regardless of billing type)
- Fix WebSocket client disconnect detection on Windows by adding
  "an established connection was aborted" to known disconnect errors

Antigravity APIKey accounts were incorrectly routed to
testAntigravityAccountConnection which calls AntigravityTokenProvider,
but the token provider only handles OAuth and Upstream types, causing
"not an antigravity oauth account" error.

Extract routeAntigravityTest to route APIKey accounts to native
Claude/Gemini test paths based on model prefix, matching the
gateway_handler routing logic for normal requests.

- Add configurable spending limit (quota_limit) for apikey-type accounts
- Atomic quota accumulation via PostgreSQL JSONB operations on TotalCost
- Scheduler filters out over-quota accounts with outbox-triggered snapshot refresh
- Display quota usage ($used / $limit) in account capacity column
- Add "Reset Quota" action in account menu to reset usage to zero
- Editing account settings preserves quota_used (no accidental reset)
- Covers all 3 billing paths: Anthropic, Gemini, OpenAI RecordUsage

chore: bump version to 0.1.90.4

When FetchUsageWithOptions receives a non-200 response from the
Anthropic API (e.g. 429 Rate Limited, 401 Unauthorized), the error
was wrapped with fmt.Errorf which infraerrors.FromError cannot
recognize, causing a generic "internal error" message with no details.

Replace fmt.Errorf with infraerrors.New(500, "UPSTREAM_ERROR", msg)
so the upstream error details (status code + body) are included in
the 500 response message. The HTTP status remains 500 to avoid
interfering with frontend auth routing (e.g. 401 would trigger
JWT expiry redirect).

- 移除 ScheduledTestOutcome 中间类型，RunTestBackground 直接返回 *ScheduledTestResult
- 简化 SaveResult 直接接受 *ScheduledTestResult
- 移除 handler 中不必要的 nil 检查
- 移除前端 ScheduledTestsPanel 中多余的 String() 转换
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

将 redis leader lock 逻辑从 service 层抽取为 LeaderLocker 接口，
实现移至 repository 层，消除 service 层对 redis 的直接依赖。
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

每个测试计划绑定一个账号和一个模型，按 cron 表达式定期执行测试，
保存历史结果并在前端账号管理页面中提供完整的增删改查和结果查看功能。

主要变更：
- 新增 scheduled_test_plans / scheduled_test_results 两张表及迁移
- 后端 service 层：CRUD 服务 + 后台 cron runner（每分钟扫描到期计划并发执行）
- RunTestBackground 方法通过 httptest 在内存中执行账号测试并解析 SSE 输出
- Redis leader lock + pg_try_advisory_lock 双重保障多实例部署只执行一次
- REST API：5 个管理端点（计划 CRUD + 结果查询）
- 前端 ScheduledTestsPanel 组件：计划管理、启用开关、内联编辑、结果展开查看
- 中英文 i18n 支持
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

新增 OpenAI WebSocket v2 passthrough relay 数据面与服务适配层，
支持按账号 ws mode 在 ctx_pool 与 passthrough 间路由。

同步调整前端 OpenAI ws mode 选项为 off/ctx_pool/passthrough，
并补充 i18n 文案与对应单测。

新增 Caddyfile.dmit 与 docker-compose-aicodex.yml 部署配置，
用于宿主机场景下的反向代理与服务编排。
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>