Skip to content

gh-v3.0.15rc4

Pre-release
Pre-release

Choose a tag to compare

View all tags
@github-actions github-actions released this 05 Sep 05:07
· 22 commits to master since this release
gh-v3.0.15rc4
b4215a0
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Highlights

  • OpenCode Model Hub is now more predictable across protocols: you can choose OpenAI Responses or Anthropic Messages per model, Gateway mode always prepares the required overlay before launch, and invalid derived state is rebuilt automatically.
  • Avibe now manages CLIProxyAPI (CPA) as a required dependency. Settings shows its installed and target versions with install, update, and repair actions, while install, upgrade, and startup automatically reconcile it even when Model Hub is disabled.
  • GPT-6 Astra is now the top built-in Codex model choice, with the full native reasoning ladder from low through ultra.
  • Agent behavior is more reliable across backends: Codex once again receives Avibe’s quick-reply and session-title instructions, while managed Skills only route when relevant Skill content is available.

Changes

Added

  • Added release-pinned CPA dependency management and lifecycle reconciliation. (#1862)
  • Added GPT-6 Astra metadata and its low, medium, high, xhigh, max, and ultra reasoning tiers to the Codex catalog. (#1864)
  • Added an OpenCode model editor API selector for OpenAI Responses and Anthropic Messages. (#1865, #1867)

Changed

  • Switched OpenCode Model Hub entries to canonical bare model IDs, persisted each model’s native protocol, and generated protocol-specific providers and overlays. (#1865)
  • Stabilized generated capability guidance by keeping required routing intact and omitting unavailable or empty prompt sections. (#1863)

Fixed

  • Prevented Gateway mode from launching OpenCode before its Model Hub overlay is prepared. (#1870)
  • Quarantined unreadable Model Hub sources.json state and rebuilt it from the current configuration. (#1869)
  • Preserved catalog-declared ultra reasoning settings when saving and reading OpenCode configuration. (#1872)
  • Restored model-visible Avibe instructions in Codex sessions without changing existing thread identity or model routing. (#1875)
  • Enforced managed Skill routing across Claude, Codex, and OpenCode, including blocking Claude’s native Skill path. (#1866)
  • Displayed Memory browse subjects as card titles, with summary and body fallbacks. (#1868)
  • Improved internal Model Hub test type-checking and aligned fixtures with contract version 8. (#1871, #1873)

Full Changelog: gh-v3.0.15rc3...gh-v3.0.15rc4

亮点

  • OpenCode Model Hub 的协议处理更稳了：每个模型都可以选择 OpenAI Responses 或 Anthropic Messages；Gateway 模式会先准备好 overlay 再启动 OpenCode，派生状态损坏时也能自动重建。
  • Avibe 现在会把 CLIProxyAPI（CPA） 作为必需依赖统一管理。依赖设置页会显示当前版本和目标版本，并提供安装、更新、修复操作；即使关闭 Model Hub，安装、升级和启动时仍会自动校准 CPA。
  • Codex 内置模型列表现在首选 GPT-6 Astra，并支持从 low 到 ultra 的完整原生推理档位。
  • 多后端下的 Agent 行为更可靠：Codex 能重新收到 Avibe 的快捷回复和会话标题指令，托管 Skill 也只会在相关内容确实存在时触发路由。

变更

新增

  • 加入按发行版本锁定的 CPA 依赖管理，并在安装、升级和启动流程中自动校准。 (#1862)
  • 在 Codex 模型目录中加入 GPT-6 Astra，以及 low、medium、high、xhigh、max、ultra 推理档位。 (#1864)
  • 为 OpenCode 模型编辑器加入 API 选择项，可选 OpenAI Responses 或 Anthropic Messages。 (#1865, #1867)

调整

  • OpenCode Model Hub 改用规范化的纯模型 ID，持久化模型原生协议，并按协议生成独立的 provider 和 overlay。 (#1865)
  • 稳定能力提示词的生成逻辑，保留必要路由，并跳过不可用或内容为空的提示区块。 (#1863)

修复

  • 修复 Gateway 模式可能在 Model Hub overlay 准备完成前启动 OpenCode 的问题。 (#1870)
  • 遇到无法读取的 Model Hub sources.json 时先隔离旧状态，再根据当前配置自动重建。 (#1869)
  • 修复 OpenCode 配置保存或读取时丢失目录声明的 ultra 推理档位。 (#1872)
  • 恢复 Codex 会话中模型可见的 Avibe 指令，同时保持已有会话 ID 和模型路由不变。 (#1875)
  • 统一 Claude、Codex 和 OpenCode 的托管 Skill 路由，并阻止 Claude 绕过托管流程调用原生 Skill。 (#1866)
  • Memory 浏览卡片优先显示 subject；缺失时依次回退到 summary 和正文。 (#1868)
  • 完善 Model Hub 测试项目的类型检查，并将测试夹具同步到合约版本 8。 (#1871, #1873)

Full Changelog: gh-v3.0.15rc3...gh-v3.0.15rc4

Assets 17
Loading