Pi v0.56.2 - 更新日志

### 新功能 - 在 `openai`、`openai-codex`、`azure-openai-responses` 和 `opencode` 中全面支持 GPT-5.4，其中 `gpt-5.4` 现已成为 `openai` 和 `openai-codex` 的默认模型 ([README.md](README.md), [docs/providers.md](docs/providers.md))。 - 新增 `treeFilterMode` 设置项，用于选择默认的 `/tree` 过滤模式（`default`、`no-tools`、`user-only`、`labeled-only`、`all`） ([docs/settings.md](docs/settings.md), [#1852](https://github.com/badlogic/pi-mono/pull/1852) by [@lajarre](https://github.com/lajarre))。 - Mistral 原生对话集成，采用基于 SDK 的 Provider 行为，保留 Mistral 特有的思考与重放语义 ([README.md](README.md), [docs/providers.md](docs/providers.md), [#1716](https://github.com/badlogic/pi-mono/issues/1716))。 ### 新增 - 为 `openai`、`openai-codex`、`azure-openai-responses` 和 `opencode` 提供商新增 `gpt-5.4` 模型支持。 - 为 `github-copilot` 新增 `gpt-5.3-codex` 备用模型支持，直至上游模型目录收录该模型 ([#1853](https://github.com/badlogic/pi-mono/issues/1853))。 - 新增 `treeFilterMode` 设置项，用于选择默认的 `/tree` 过滤模式（`default`、`no-tools`、`user-only`、`labeled-only`、`all`） ([#1852](https://github.com/badlogic/pi-mono/pull/1852) by [@lajarre](https://github.com/lajarre))。 ### 变更 - 将 `openai` 和 `openai-codex` 提供商的默认模型更新为 `gpt-5.4`。 ### 修复 - 修复了 GPT-5.3 Codex 后续对话轮次丢失 OpenAI Responses 助手 `phase` 元数据的问题：通过在会话历史中保留可重放签名，并将 `phase` 转发回 Responses API ([#1819](https://github.com/badlogic/pi-mono/issues/1819))。 - 修复了 OpenAI Responses 重放逻辑，跳过空思考块，避免在后续对话轮次中产生无效的空操作推理项。 - 更新 Mistral 集成，改用基于原生 SDK 的 Provider 和对话 API，包括 coding-agent 模型/Provider 的关联配置及 Mistral 设置文档 ([#1716](https://github.com/badlogic/pi-mono/issues/1716))。 - 提升 Antigravity 可靠性：处理 403/404 时的端点级联问题，新增 autopush 沙箱回退机制，移除多余的指纹请求头 ([#1830](https://github.com/badlogic/pi-mono/issues/1830))。 - 修复已发布安装中 `@mariozechner/pi-ai/oauth` 扩展的导入问题：改为直接从构建好的 `dist` 文件解析子路径，而非使用包根目录的包装垫片 ([#1856](https://github.com/badlogic/pi-mono/issues/1856))。 - 修复 Gemini 3 多轮工具调用丢失结构化上下文的问题：对未签名的函数调用使用 `skip_thought_signature_validator` 哨兵值，而非回退为文本 ([#1829](https://github.com/badlogic/pi-mono/issues/1829))。 - 修复了模型选择器过滤框在 VS Code 1.110+ 中无法接受输入字符的问题：原因是 `Input` 组件缺少 Kitty CSI-u 可打印字符解码 ([#1857](https://github.com/badlogic/pi-mono/issues/1857))。 - 修复了终端调整大小时编辑器/页脚可见性漂移的问题：在终端宽度或高度变化时强制进行完整重绘 ([#1844](https://github.com/badlogic/pi-mono/pull/1844) by [@ghoulr](https://github.com/ghoulr))。 - 修复了页脚宽度对宽 Unicode 文本（会话名称、模型、提供商）的截断问题，防止渲染行超出终端宽度导致 TUI 崩溃 ([#1833](https://github.com/badlogic/pi-mono/issues/1833))。 - 修复了 Windows 下预览背景出现的伪影问题：在工具输出预览的显示渲染中，将 CRLF 内容 (` `) 规范化为 LF ([#1854](https://github.com/badlogic/pi-mono/issues/1854))。

### New Features - GPT-5.4 support across `openai`, `openai-codex`, `azure-openai-responses`, and `opencode`, with `gpt-5.4` now the default for `openai` and `openai-codex` ([README.md](README.md), [docs/providers.md](docs/providers.md)). - `treeFilterMode` setting to choose the default `/tree` filter mode (`default`, `no-tools`, `user-only`, `labeled-only`, `all`) ([docs/settings.md](docs/settings.md), [#1852](https://github.com/badlogic/pi-mono/pull/1852) by [@lajarre](https://github.com/lajarre)). - Mistral native conversations integration with SDK-backed provider behavior, preserving Mistral-specific thinking and replay semantics ([README.md](README.md), [docs/providers.md](docs/providers.md), [#1716](https://github.com/badlogic/pi-mono/issues/1716)). ### Added - Added `gpt-5.4` model availability for `openai`, `openai-codex`, `azure-openai-responses`, and `opencode` providers. - Added `gpt-5.3-codex` fallback model availability for `github-copilot` until upstream model catalogs include it ([#1853](https://github.com/badlogic/pi-mono/issues/1853)). - Added `treeFilterMode` setting to choose the default `/tree` filter mode (`default`, `no-tools`, `user-only`, `labeled-only`, `all`) ([#1852](https://github.com/badlogic/pi-mono/pull/1852) by [@lajarre](https://github.com/lajarre)). ### Changed - Updated the default models for the `openai` and `openai-codex` providers to `gpt-5.4`. ### Fixed - Fixed GPT-5.3 Codex follow-up turns dropping OpenAI Responses assistant `phase` metadata by preserving replayable signatures in session history and forwarding `phase` back to the Responses API ([#1819](https://github.com/badlogic/pi-mono/issues/1819)). - Fixed OpenAI Responses replay to omit empty thinking blocks, avoiding invalid no-op reasoning items in follow-up turns. - Updated Mistral integration to use the native SDK-backed provider and conversations API, including coding-agent model/provider wiring and Mistral setup documentation ([#1716](https://github.com/badlogic/pi-mono/issues/1716)). - Fixed Antigravity reliability: endpoint cascade on 403/404, added autopush sandbox fallback, removed extra fingerprint headers ([#1830](https://github.com/badlogic/pi-mono/issues/1830)). - Fixed `@mariozechner/pi-ai/oauth` extension imports in published installs by resolving the subpath directly from built `dist` files instead of package-root wrapper shims ([#1856](https://github.com/badlogic/pi-mono/issues/1856)). - Fixed Gemini 3 multi-turn tool use losing structured context by using `skip_thought_signature_validator` sentinel for unsigned function calls instead of text fallback ([#1829](https://github.com/badlogic/pi-mono/issues/1829)). - Fixed model selector filter not accepting typed characters in VS Code 1.110+ due to missing Kitty CSI-u printable decoding in the `Input` component ([#1857](https://github.com/badlogic/pi-mono/issues/1857)) - Fixed editor/footer visibility drift during terminal resize by forcing full redraws when terminal width or height changes ([#1844](https://github.com/badlogic/pi-mono/pull/1844) by [@ghoulr](https://github.com/ghoulr)). - Fixed footer width truncation for wide Unicode text (session name, model, provider) to prevent TUI crashes from rendered lines exceeding terminal width ([#1833](https://github.com/badlogic/pi-mono/issues/1833)). - Fixed Windows write preview background artifacts by normalizing CRLF content (`\r\n`) to LF for display rendering in tool output previews ([#1854](https://github.com/badlogic/pi-mono/issues/1854)).