Commit Graph
3654 Commits
Author SHA1 Message Date
alfadb 0aaed397cc fix: preserve Astra reasoning mode for Codex OAuth 2026-09-07 17:58:20 +08:00
Wesley Liddick b8ff74ad70 Merge pull request #6668 from ssharkkky/fix/channel-monitor-gemini-role
fix(monitors): send explicit user role in Gemini monitor probes
2026-09-07 17:38:00 +08:00
Wesley Liddick 96cf8bd76a Merge pull request #6665 from ssharkkky/fix/grok-cross-client-map-default-doc
fix(settings): describe Grok cross-client mapping default as enabled
2026-09-07 17:37:47 +08:00
Wesley Liddick 787a6a33df Merge pull request #6481 from alfadb/fix/claude-cli-version-floor
fix(claude): 让伪装的 Claude Code 版本真正生效,修复 Fable 5.1 仍然 400
2026-09-07 17:37:12 +08:00
Wesley Liddick f8e1e7fed6 Merge pull request #6711 from wucm667/fix/issue-6701-model-not-found-failover
Fix failover for model-not-found responses
2026-09-07 17:36:58 +08:00
Wesley Liddick f09a5b602e Merge pull request #6704 from aofee/codex/fix-http-bridge-session-isolation
fix(openai-ws): 修复 HTTP bridge 并发连接抢占与状态串用
2026-09-07 17:36:36 +08:00
Wesley Liddick 027e3f2d54 Merge pull request #6688 from suixincrazy/fix/codex-manifest-failover-alias
fix(openai): honor model_routing for Codex manifests and scheduling
2026-09-07 17:36:26 +08:00
Wesley Liddick 959cbe3d87 Merge pull request #5453 from wucm667/fix/issue-5451-ws-passthrough-before-turn
fix(openai): reacquire passthrough turn slots
2026-09-07 17:35:57 +08:00
Wesley Liddick c0420e2b8a Merge pull request #6484 from aofee/codex/fix-fable-credits-required-scope
fix(anthropic): keep Fable credits_required model-scoped
2026-09-07 17:33:59 +08:00
Wesley Liddick 9578ddbddd Merge pull request #6687 from Chenning-Tao/codex/fix-native-delegation-ws
fix(openai): preserve named standalone Responses inputs
2026-09-07 17:31:17 +08:00
Wesley Liddick e274de45b7 Merge pull request #6702 from a1gaoshanlaiyige/fix/message-cache-json-escaping
fix(gateway): use JSON escaping for message cache promotion
2026-09-07 17:30:54 +08:00
Wesley Liddick ce328fb37a Merge pull request #6670 from ssharkkky/fix/payment-fulfillment-isolation
fix(payment): isolate fulfillment from redeem limits
2026-09-07 17:30:43 +08:00
alfadb ce157b32ed fix(claude): 指纹自愈路径同样执行 claude-cli 版本下限
GetOrCreateFingerprint 的 heal 分支(畸形缓存自愈)此前在版本下限逻辑之外:
畸形缓存 + 合法但过旧的客户端 UA 会在首次读取时把过旧 UA 原样落库并返回,
下一次读取才被抬到下限——该窗口内账号仍以过旧版本撞上游客户端版本闸门
(PR #6481 review 指出的边界 case)。

修复:把 floorClaudeCLIUserAgentVersion 后置块从 healthy else 内移到
heal/healthy if-else 之后、TTL 续期检查之前,作为两条缓存命中路径的共同
后置条件;heal 分支 needWrite 本已为 true,单次落库语义不变。

测试:TestGetOrCreateFingerprintHealsPoisonedCacheUsingValidClientUA
期望改为首次返回/落库即 claude-cli/CLICurrentVersion(保留 ClientID 不变、
恰好一次落库断言);floor/UA 全部既有用例保持绿。
2026-09-07 16:34:35 +08:00
alfadb 3cb2381bdc fix(claude): 让伪装的 Claude Code 版本真正生效,修复 Fable 5.1 仍然 400
claude-fable-5-1 已登记进 DefaultModels,但 OAuth 账号实际调用仍然全部失败:

  400 claude_code_version_too_old
  "Claude Code 2.1.220 does not support this model;
   version 2.1.251 or newer is required."

两个独立的原因,缺一不可:

1. CLICurrentVersion 停留在 2.1.220,低于上游对该模型的客户端版本闸门。
   升到 2.1.258(当前官方 CLI 最新版,npm @anthropic-ai/claude-code)。

2. 只升常量对所有已有账号无效。上游请求实际发送的 User-Agent 来自账号级
   持久指纹(IdentityService,Redis fingerprint:<accountID>),不是
   DefaultHeaders;defaultFingerprint 只在首次创建指纹时使用。
   GetOrCreateFingerprint 的缓存命中路径只有两条改写分支:UA 畸形时自愈,
   或客户端送来更新版本时升级。指纹 7 天 TTL 但活跃账号每 24h 续期,实际
   近乎永不过期,系统内也没有重置入口——客户端版本比缓存旧时(例如缓存
   2.1.220、客户端 2.1.75),存量账号会永远对上游声称 2.1.220。

   生产实测:升级 CLICurrentVersion 并部署后仍报同样的 400,读 Redis 确认
   出问题账号的缓存值为 "claude-cli/2.1.220 (external, cli)"。

修复:新增 floorClaudeCLIUserAgentVersion,把 CLICurrentVersion 作为缓存
指纹里 claude-cli 版本的下限,缓存命中与首次创建两条路径共用。

  - 只对 claude-cli 产品生效,其它客户端一律不动
  - 只升不降:版本等于或高于下限(含客户端上报的更新版本)不做任何改动
  - 只替换版本号段,UA 其余部分原样保留,例如
    "(external, claude-desktop-3p, agent-sdk/0.3.100)" 不受影响
  - 与既有的"客户端更新版本"升级分支相互独立、二者取更新者
  - X-Stainless-* 维持既有 merge 语义,不被下限逻辑覆盖

cc_version 无需单独处理:billing attribution block 的版本已由
syncBillingHeaderVersion 跟随实际发送的 User-Agent 同步。

测试:identity_service_version_floor_test.go 覆盖 helper 的 8 种 UA 形态与
5 个 GetOrCreateFingerprint 场景,其中一个直接复现线上故障(缓存 2.1.220 +
客户端 2.1.75 → 写回 2.1.258,Stainless 字段保持不变)。
constants_cli_version_test.go 锁定常量与 DefaultHeaders 的 UA 同源,并断言
不低于 Fable 5.1 的闸门下限。回退实现后对应用例确认转红。

identity_service_user_agent_validation_test.go 中 3 个把版本钉死在旧字面量
的既有用例改为钉 CLICurrentVersion 或更高版本——它们断言的正是本次有意改变
的行为,各自的原始意图(哨兵拒绝、合法升级生效、健康缓存不重写)保持不变。
2026-09-07 16:34:35 +08:00
Wesley Liddick a7d2c782df Merge pull request #5717 from wucm667/feat/issue-2629-current-main
feat(simple-mode): allow basic account grouping
2026-09-07 15:43:22 +08:00
wucm667 b59f3bf46c fix(simple-mode): enforce basic group boundaries 2026-09-07 09:20:48 +08:00
Zero Clover f88d62ad28 fix(openai): use live account models in connection tests 2026-09-07 00:23:49 +08:00
wucm667 8363d537e0 fix(gateway): fail over model-not-found responses 2026-09-06 20:38:17 +08:00
aofee 96884dd1ae fix(openai-ws): isolate HTTP bridge connection state 2026-09-06 17:05:06 +08:00
a1gaoshanlaiyige 0b061c5fa1 fix(gateway): use JSON escaping for message cache promotion
Use encoding/json instead of Go string quoting when promoting message content to cached text blocks. Add regression coverage for control characters, Unicode, literal backslashes, text preservation, and idempotence.
2026-09-06 16:42:57 +08:00
suixincrazy 28167bbf85 fix(openai): honor model_routing for Codex manifests and scheduling
An OpenAI group that keeps a failover account for a shared public alias
produced a Codex model descriptor with no reasoning levels and text-only
input, which leaves the client's reasoning-effort slider with nothing to
select.

Two accounts explicitly mapping one alias to different upstream models
makes codexExplicitModelTargetsConflictForPlatform true. With no
upstream_model_metadata snapshot to intersect, groupCodexModelMetadata
short-circuits to {reasoningConflict, inputModalitiesConflict}, so
DefaultReasoningLevel is nil (the key vanishes under omitempty) and
SupportedReasoningLevels is an empty array. mergeMissingCodexModelFields
fills only missing-or-null fields, so an empty array is never repaired.

Failing closed is right when the alias target is genuinely ambiguous, but
a group that declares model_routing for the alias has already resolved
that ambiguity: the operator stated which accounts serve it and which is
primary. Treating that as a conflict rejects a deliberate failover setup.

Manifest side: when a routing rule names the alias, restrict capability
declaration to the routed accounts and skip the conflict override.
Capabilities then fall back to the same by-name derivation used for a
single account without a metadata snapshot. Groups with no rule, a
disabled rule, or a rule for another alias keep failing closed.

Scheduling side: model_routing was stored and read platform-agnostically
(admin_group.go, Group.GetRoutingAccountIDs) but four gates in
gateway_scheduling.go only consulted it for Anthropic, so an OpenAI
group's rules were accepted and displayed yet never honored. Without
this, the manifest would honor a rule that scheduling ignores. Both
gates now share one predicate, with the allowed platform set defined in
a single place.

Selection inside a routing list is unchanged: the list is a membership
set, and the winner is still decided by Priority, then load, then
LastUsedAt.

Tests: alias-failover manifest recovery plus the three negative cases,
the platform predicate table, and an OpenAI routed selection that must
beat a higher-priority unrouted account.
2026-09-06 10:12:20 +08:00
Chenning-Tao 985428d9d2 fix(openai): preserve named standalone Responses inputs 2026-09-06 09:33:53 +08:00
wucm667 4a4fd35e0c fix(simple-mode): guard non-empty group deletion 2026-09-06 08:11:23 +08:00
TT 7a70de401c fix(payment): isolate fulfillment from redeem limits
Payment and admin fulfillment shared the public Redeem entry point, so
fulfillment retries were blocked by, and contributed to, the user's
public redeem failure counter, and lookup errors were silently treated as
"code does not exist" on a partial re-run.

- resolveRedeemAction now distinguishes a genuine not-found lookup from
  other lookup errors (which now fail the fulfillment instead of being
  swallowed and re-created).
- Add validatePaymentRedeemCode: before reusing an existing code during
  fulfillment, verify it belongs to the order (code id, balance type,
  amount, user, status), so a stale or foreign code can never be
  credited against the wrong order/user.
- Split Redeem into a policy-based core: public redemption enforces the
  failure counter; payment fulfillment (redeemForPaymentFulfillment) and
  admin fulfillment (RedeemForAdminFulfillment) bypass it, while all code
  validation and transactional updates remain identical. Admin
  fulfillment keeps the normal redeem-level affiliate rebate.
- The admin CreateAndRedeem endpoint now uses RedeemForAdminFulfillment.

Tests: updated resolveRedeemAction cases, new TestValidatePaymentRedeemCode,
code-used-by-another-user rejection, payment/admin rate-limit bypass
coverage, and public-path regression coverage.
2026-09-05 23:56:59 +08:00
TT c7343d2aa2 fix(monitors): send explicit user role in Gemini monitor probes
The Antigravity native Gemini route (generateContent) expects each
contents entry to carry an explicit "user" role, matching the request
shape used by the established Antigravity account-test flow. Without it
the probe body is rejected with 400 INVALID_ARGUMENT, so Gemini channel
monitors report false failures.

Add the explicit role to the Gemini provider adapter body and cover it
with a regression test that parses the generated request body.
2026-09-05 23:30:45 +08:00
TT 32bf3d0f33 fix(settings): describe Grok cross-client mapping default as enabled
The Grok cross-client model mapping is enabled by default: InitializeDefaultSettings writes "true", and the parser treats a missing/empty value as enabled ("Default true (missing/empty -> enabled)"). However the comment above the defaults and the admin settings UI hint (EN and ZH) both claimed "disabled by default", contradicting the actual behavior.

Align the comment and both UI hints with the real default. No behavior change.
2026-09-05 23:17:34 +08:00
Zero Clover cb31033972 feat(openai): reuse pinned accounts for standard model discovery 2026-09-05 19:58:47 +08:00
Wesley Liddick 578785ee7f Merge pull request #6640 from cyhhao/fix/claude-billing-fingerprint
fix(claude): align billing fingerprint with outbound User-Agent
2026-09-05 16:59:33 +08:00
Wesley Liddick ed7c8f2208 Merge pull request #6572 from lyy0709/codex/gpt-6-astra
feat(openai): complete GPT-6 Astra support
2026-09-05 16:50:40 +08:00
cyh 2e1c7c00a6 test(claude): check request body close errors 2026-09-05 16:43:54 +08:00
cyh 994ca26e92 fix(claude): align billing fingerprint with outbound user agent 2026-09-05 16:34:00 +08:00
Wesley Liddick d2f87c612e Merge pull request #6638 from feeeei/feat/images-url-to-b64json-backfill
feat(images): 账户开关将生图结果 url 下载回填为 b64_json
2026-09-05 16:24:40 +08:00
lyy0709 62cd63bfda fix(openai): retain capabilities when model listing is unavailable 2026-09-05 16:24:15 +08:00
lyy0709 dee35a841d merge: reconcile Astra fixes with upstream capability sync 2026-09-05 16:19:57 +08:00
Wesley Liddick fb44e2dd09 Merge pull request #6514 from StarryKira/codex/anthropic-reasoning-pricing
feat: add Anthropic reasoning effort and Fable 5.1 max pricing
2026-09-05 16:19:18 +08:00
Wesley Liddick c42d78e26c Merge pull request #6571 from DuskLin/feat/codex-ultrafast-tier
feat(codex): support ultrafast service tier
2026-09-05 16:18:16 +08:00
Wesley Liddick 2dc287f2e0 Merge pull request #6636 from specialpointcentral/cyber-ws-recording-fix
fix(openai): record WS cyber policy failures across paths
2026-09-05 16:17:33 +08:00
Wesley Liddick 83094abf21 Merge pull request #6536 from wucm667/fix/issue-6526-mapped-model-scheduling
fix(openai): schedule accounts with mapped model
2026-09-05 16:17:04 +08:00
lyy0709 d7f048a265 fix(openai): preserve synced Astra capabilities and continuation 2026-09-05 16:13:28 +08:00
Wesley Liddick 0d9e7e1530 Merge pull request #6628 from baryon/contrib/openai-gpt6-astra-capability-sync
fix(openai): persist GPT-6 Astra Codex capabilities from upstream sync
2026-09-05 16:07:50 +08:00
Wesley Liddick 9b4fcfb897 Merge pull request #6606 from luckydududu/feat/claude-cli-version-env-override
feat(claude): allow operators to override the advertised CLI version
2026-09-05 16:05:33 +08:00
Wesley Liddick 9517f12cdd Merge pull request #6510 from wall-wxk/fix/release-session-slot-on-forward-failure
fix(gateway): 转发失败的请求立即释放会话槽,不再占用整个空闲超时窗口
2026-09-05 16:05:06 +08:00
Wesley Liddick 40f4d450be Merge pull request #6507 from wucm667/fix/issue-6501-glm53-thinking-effort
fix(gateway): normalize GLM-5.3 thinking effort
2026-09-05 16:04:16 +08:00
Wesley Liddick 620eb3fd00 Merge pull request #6581 from r266-tech/r266/forward-opencode-session
fix(openai): scope OpenCode session forwarding
2026-09-05 16:03:57 +08:00
Wesley Liddick 7b271bbee9 Merge pull request #6529 from wucm667/fix/issue-6521-preserve-passthrough-reasoning
fix(openai): preserve reasoning effort in API-key passthrough
2026-09-05 16:02:56 +08:00
feeeei c227863d51 fix(images): 生图 url 回填下载拒绝私网目的地并只信字节嗅探
账户开关 images_url_to_b64_json 下载的图片 url 来自上游响应,是不可信输入。此前沿用
base_url 的 security.url_allowlist 策略,默认配置下只做格式校验,上游返回
localhost:8080 之类地址时网关会真的去请求内网。

- 下载 URL 主机为 localhost、回环、私网、链路本地、未指定地址的字面量时直接拒绝,
  不受 security.url_allowlist 配置影响(urlvalidator 导出 IsBlockedHost)
- 新增 ctx 标记 WithHTTPUpstreamPublicHostsOnly:共享上游客户端对带标记的请求无视配置
  校验解析后 IP,并为其克隆客户端挂 redirectChecker,重定向保持跟随但每一跳都校验
- 下载内容改为只信 http.DetectContentType 嗅探结果且限 png/jpeg/webp/gif,
  响应头不再作为依据
- 下载失败仍只记日志并保留原样;超时与端口策略不变
2026-09-05 15:50:56 +08:00
lyy0709 3a8c9f16de Merge remote-tracking branch 'upstream/main' into codex/gpt-6-astra 2026-09-05 15:48:49 +08:00
Qi HU 2da31290a1 fix(openai): record WS cyber policy failures across paths
Cyber-policy detection only inspected response.failed, and the ingress check ran after error-specific early returns. As a result, error-shaped failures and the connection-pool, HTTP-bridge, and WSv2 paths could miss cyber-policy hits.

Centralize event marking and invoke it before retry and failover returns across the WebSocket implementations. Preserve the recorded guard during account failover and defer connection blocking until the logical failover chain completes, avoiding duplicate records while ensuring subsequent turns are blocked.
2026-09-05 15:32:29 +08:00
feeeei 3c53ba01a7 feat(images): 账户开关将生图结果 url 下载回填为 b64_json
部分第三方 GPT-image 上游把 Images 响应里的 b64_json 换成 url,按官方接口实现的
客户端读不到图。账户 extra.images_url_to_b64_json 开启后,OpenAI API Key 的
Images 非流式响应里 b64_json 缺失或为空且 url 非空的项,由网关下载 url 内容并以
base64 回填 b64_json,url 与其余字段原样保留;异步生图任务复用同一路径。

- 下载经账户代理发起,沿用 base_url 的出站 URL 校验策略(validateUpstreamBaseURL
  抽出 validateOutboundURL,nil cfg 按仅格式校验处理),单图上限 20MB 且须为图片
  内容(响应头或字节嗅探);data: URL 直接取 base64 载荷
- 单项下载失败仅记日志并保留原样;客户端显式 response_format=url 时不回填;
  流式响应不处理
- 创建/编辑账户弹窗在 OpenAI API Key 的 Responses 模式块后新增“生图结果 URL 转
  base64”开关,中英文案;新增 openai_images_b64_backfill_test.go
- 只被 unit 标签测试引用的定价目录测试辅助补上 //go:build unit(admin 包整文件加标签,
  service 包的 newPricingServiceFromModelPricing 搬到 pricing_stub_helpers_unit_test.go),
  消除 golangci-lint 在默认标签下的 unused 报错
2026-09-05 15:17:57 +08:00
Wesley Liddick 07bf8b92fd Merge pull request #6550 from weiness/fix/alipay-payment-reconciliation
修复支付宝回调延迟或丢失,已支付订单可能长时间停留在待支付状态的问题
2026-09-05 15:08:56 +08:00