Skip to content

RELEASE.md

Latest commit

 

History

History
95 lines (59 loc) · 6.14 KB

RELEASE.md

File metadata and controls

95 lines (59 loc) · 6.14 KB
Raw

StackFerry v0.1.27

修复 Codex 中 GPT-6-Astra 需要先拉取模型才出现在菜单的问题。

简体中文

Codex GPT-6-Astra

  • 未配置显式模型目录的 GPT 兼容配置会自动生成包含当前模型和 gpt-6-astra 的目录，无需先请求供应商 /models；带命名空间的模型保留原有前缀。
  • 修复旧模板中的隐藏标记导致模型无法出现在 Codex /model 菜单的问题，并保留 Astra 的六档推理强度（low、medium、high、xhigh、max、ultra）及 872K 最大上下文元数据。
  • 本地路由接管时区分“未配置目录”和“显式空目录”：前者生成默认目录，后者仍清除旧目录；已有自定义映射和非 GPT 模型不自动加入 Astra。
  • 新增三项定向回归测试，覆盖离线菜单、推理强度、显式映射和接管状态。本次未修改官方文档 MCP 的地址或鉴权逻辑，也不包含其间歇性 403 的修复。

模型目录可见性不代表供应商必定支持该模型；实际请求仍取决于上游支持。更新后需让 StackFerry 同步 Codex 配置，并重启已有 Codex 会话以加载目录。

English

Fix GPT-6-Astra requiring a model fetch before appearing in Codex's picker.

Codex GPT-6-Astra

  • GPT-compatible configurations without an explicit model catalog now generate a catalog containing the active model and gpt-6-astra without fetching the provider's /models endpoint. Namespaced models keep their prefix.
  • Override stale hidden-template flags so catalog models appear in Codex's /model picker, preserving Astra's six reasoning levels (low, medium, high, xhigh, max, ultra) and 872K maximum context metadata.
  • Local routing takeover now distinguishes missing catalogs from explicitly empty catalogs: missing catalogs allow defaults, while empty catalogs still clear old entries. Explicit mappings and non-GPT models do not receive an added Astra entry.
  • Added three targeted regression tests for offline picker entries, reasoning levels, explicit mappings, and takeover state. This release does not change the official documentation MCP endpoint or authentication and does not fix its intermittent HTTP 403 responses.

Picker visibility does not guarantee upstream model support. After updating, let StackFerry synchronize the Codex configuration and restart existing Codex sessions to load the catalog.

Previous release

自 v0.1.24 发布以来的更新（并包含此前未完成发布的故障转移修复）。

简体中文

发布与 CI

  • 补齐故障转移运行态控制的发布流程，生成新的跨平台安装包和自动更新清单。
  • 修复 ProviderList 测试 mock 与格式检查，确保前端 CI 检查通过。

产品公告

  • 增加 StackFerry 重构期间的维护公告，明确现有稳定版本、配置和供应商路由仍可继续使用。

故障转移渠道控制

  • 故障转移队列新增按渠道控制的启用/禁用开关。禁用渠道保留原有 P 序号，但会从所有故障转移候选中跳过。
  • 渠道状态使用事务化更新，保护最后一个启用渠道，并在重新加入或清空队列时正确恢复默认启用状态。
  • Provider 卡片与设置页故障转移队列使用同一套状态 mutation，避免两个入口产生不同结果。

实际运行渠道与状态展示

  • 新增按应用隔离的运行时渠道跟踪，只在主请求完整成功后显示实际使用的 Provider。
  • 修复故障转移成功后仍显示配置渠道的问题；Codex 不会因运行时切换改写持久化配置或 live 配置。
  • Pi、标准路由、图片生成、联网搜索、激活选择和批量检查统一过滤禁用渠道；辅助请求不会覆盖主请求当前渠道。
  • 故障转移、接管关闭、渠道移除或禁用后会立即清理失效的运行时状态，并通过事件刷新界面。

数据库与质量

  • 数据库 schema 升级到 v30，为既有 Provider 增加故障转移渠道参与状态并兼容旧数据库。
  • SQL 备份恢复、CCSwitch 导入和普通 Provider 保存不会覆盖用户手动设置的禁用状态。
  • 增加队列顺序、并发切换、运行时状态、Pi/Codex 路由和辅助请求隔离等回归测试。

English

Changes since the v0.1.24 release, including the failover fixes whose previous release did not complete.

Release and CI

  • Completed the release path for failover runtime controls, producing new cross-platform installers and updater metadata.
  • Fixed the ProviderList test mock and formatting checks so the frontend CI scope passes.

Product Announcement

  • Added a maintenance announcement for the ongoing StackFerry rebuild, clarifying that the current stable version, configuration, and provider routes remain usable.

Failover Channel Control

  • Added a per-channel enable/disable switch to the failover queue. Disabled channels keep their P-order but are skipped by every failover candidate path.
  • Channel changes are transactional, protect the last enabled channel, and restore the default enabled state when a provider is re-added or the queue is cleared.
  • Provider cards and the settings queue manager now share the same mutation and cache behavior.

Runtime Provider State

  • Added an app-isolated runtime route tracker that marks a provider current only after a complete successful main request.
  • Fixed failover status reporting so the UI shows the provider actually serving traffic; Codex runtime failover does not rewrite persisted or live configuration.
  • Pi, standard routing, image generation, web search, activation selection, and batch checks consistently skip disabled channels, while auxiliary requests cannot overwrite the main request channel.
  • Runtime state is cleared immediately when failover, takeover, or channel membership changes invalidate it, with event-driven UI refreshes.

Database and Quality

  • Upgraded the database schema to v30 with backward-compatible failover participation state for existing providers.
  • SQL backup/restore, CCSwitch imports, and normal provider saves preserve a user-selected disabled state.
  • Added regression coverage for queue ordering, concurrent toggles, runtime state, Pi/Codex routing, and auxiliary request isolation.