Cmochance · Cmochance · May 20, 2026 · May 20, 2026 · May 20, 2026 · May 21, 2026
diff --git a/ACKNOWLEDGEMENTS.md b/ACKNOWLEDGEMENTS.md
@@ -124,8 +124,8 @@
 
 - **Link**: https://github.com/openai/codex
 - **License**: Apache-2.0
-- **借鉴形式**: Prompt 蓝本(精简移植)+ 协议反查(数据模式参照)
-- **首次借鉴 PR / 时间**: v2.0.x 起协议结构反查;fix/219 起 prompt 结构借鉴
+- **借鉴形式**: Prompt 蓝本(精简移植)+ 协议反查(数据模式参照)+ V4A apply_patch system guidance(verbatim 引用)
+- **首次借鉴 PR / 时间**: v2.0.x 起协议结构反查;fix/219 起 compact prompt 结构借鉴;PR #235 后续(post-#236)起 apply_patch V4A chat-path guidance verbatim 引用
 - **借鉴清单**:
   - `COMPACT_SUMMARIZATION_PROMPT` 基础骨架 → `crates/adapters/src/responses/compact.rs:82-92`
     (源文件:`codex-rs/core/templates/compact/prompt.md`,~460 chars)
@@ -137,6 +137,12 @@
   - `CompactHistoryResponse { output: Vec<ResponseItem> }` + `ResponseItem::Compaction { encrypted_content }` 响应结构
     → `compact.rs` 序列化路径
     (源文件:`codex-rs/codex-api/src/endpoint/compact.rs` + `codex-rs/protocol/src/models.rs:882`)
+  - apply_patch V4A chat-path system guidance(verbatim 引用)→
+    `crates/adapters/src/responses/apply_patch_v4a_reference.md`(verbatim 镜像,头部加 adapter note 显式 override "shell command" 字眼为 "function-call tool",其余原文未改动)
+    由 `crates/adapters/src/responses/request.rs::APPLY_PATCH_CHAT_PATH_SYSTEM_GUIDANCE`(`include_str!`)拼装注入
+    源文件:`codex-rs/core/prompt_with_apply_patch_instructions.md` 的 L277-L351 @ commit `0b4f86095c8005d8f74e9c62b971d72c1670aa88`
+    动机:issue #235 真机 capture 28-turn / 26MB DeepSeek+Kimi 数据,模型生成 V4A patch 在 chat-path 失败率 6 / 7 个 turn,需要在 prompt 注入完整 V4A 教学(envelope / hunks / `@@` 锚点 / 3-line context / EBNF / 多操作 example)。chat-path 无 lark grammar 兜底,只能靠 prompt 引导。
+    上游 SHA 升级时:同步更新本文件 + `apply_patch_v4a_reference.md` 头部 + `request.rs::APPLY_PATCH_CHAT_PATH_SYSTEM_GUIDANCE` doc comment + `README.md` / `README.en.md` 致谢段 + `NOTICE` 文件共 6 处 commit SHA,再 fresh slice 覆盖 reference 正文。
 - **本项目差异 / 扩展**:
   - prompt 补两条 Claude Code 关键 bullet("All user messages verbatim" + "Next Step verbatim quote"),
     借鉴自 Piebald-AI/claude-code-system-prompts 反编译公开版本第 6 / 9 段(见下方同名 entry)

diff --git a/NOTICE b/NOTICE
@@ -0,0 +1,60 @@
+Codex App Transfer
+Copyright (c) 2026 Codex App Transfer
+
+This project is licensed under the MIT License (see LICENSE.txt).
+
+=============================================================================
+Third-party attribution
+=============================================================================
+
+This project includes prose / prompt content derived from upstream open-source
+projects. The following attributions are required by the upstream licenses.
+
+-----------------------------------------------------------------------------
+openai/codex — Apache License 2.0
+-----------------------------------------------------------------------------
+
+Copyright 2025 OpenAI
+
+Licensed under the Apache License, Version 2.0 (the "License"); you may not
+use this file except in compliance with the License. You may obtain a copy
+of the License at:
+
+    https://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
+WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
+License for the specific language governing permissions and limitations
+under the License.
+
+Derived content in this repository:
+
+  * crates/adapters/src/responses/apply_patch_v4a_reference.md
+    Verbatim mirror of the V4A apply_patch walkthrough section
+    (envelope grammar / `@@` anchors / 3-line context / EBNF / multi-op
+    example) from upstream:
+      Source repo:    https://github.com/openai/codex
+      Source commit:  0b4f86095c8005d8f74e9c62b971d72c1670aa88
+      Source file:    codex-rs/core/prompt_with_apply_patch_instructions.md
+      Source lines:   L277-L351 (75 lines)
+    The mirror file adds a one-line "Adapter note" preamble (authored by
+    codex-app-transfer maintainers) clarifying that in this chat-completions
+    environment apply_patch is a function-call tool, not a shell command.
+    The borrowed prose itself is otherwise unmodified.
+
+    The mirror file is embedded into the system prompt at runtime via
+    `include_str!` from
+    `crates/adapters/src/responses/request.rs::APPLY_PATCH_CHAT_PATH_SYSTEM_GUIDANCE`.
+
+  * crates/adapters/src/responses/compact.rs (prior borrowing, retained)
+    `COMPACT_SUMMARIZATION_PROMPT` skeleton and `COMPACT_SUMMARY_PREFIX`
+    derived from
+    `codex-rs/core/templates/compact/{prompt.md,summary_prefix.md}`.
+    See ACKNOWLEDGEMENTS.md for full borrowing list and file:line mapping.
+
+-----------------------------------------------------------------------------
+
+For the complete list of upstream projects this repository borrows from
+(including non-Apache projects), see ACKNOWLEDGEMENTS.md and the credits
+section of README.md / README.en.md.
diff --git a/README.en.md b/README.en.md
@@ -41,6 +41,7 @@ With any provider enabled, Codex CLI's model picker shows `<provider> / <real-mo
 - Manage multiple providers; map OpenAI model names (`gpt-5.5` / `gpt-5.4` / `gpt-5.4-mini` / `gpt-5.3-codex` / `gpt-5.2`) to the provider's real model IDs
 - Translate Codex CLI's Responses API streaming / non-streaming requests into upstream protocols: Chat Completions, Gemini Native (`:streamGenerateContent`), Gemini CLI OAuth (Cloud Code Assist), Anthropic Messages (`/v1/messages`), Grok Web (`/rest/app-chat/conversations/new`), Responses passthrough, etc.
 - Multi-turn tool conversation context + `previous_response_id` history replay + autocompact expansion + thinking / reasoning_content injection — all aligned with the OpenAI Responses API protocol
+- Codex CLI's freeform `apply_patch` tool (edit-file +/- diff UI) works on DeepSeek / Kimi / MiMo and other chat-completions providers: the adapter bridges Responses `custom_tool_call` ↔ chat `function_call` wire forms, the model emits V4A-format patches, Codex CLI renders the diff (issue #235)
 - **Two-layer session history persistence**: L1 in-memory LRU + L2 sqlite with 30-day TTL (`~/.codex-app-transfer/sessions.db`), preserving history across `.app` restarts
 - Codex CLI config guardrails: snapshots `~/.codex/{config.toml,auth.json}` before apply; restores via per-key smart merge on exit / next start
 - Real-time logs panel auto-refreshing every 2s; unified `tracing::warn!(error_id, detail)` with stable tokens — operators can grep / aggregate
@@ -224,7 +225,7 @@ Some experimental providers (Grok Web / Gemini CLI OAuth / Antigravity OAuth) in
 - [`lonr-6/cc-desktop-switch`](https://github.com/lonr-6/cc-desktop-switch) — v1.x desktop shell skeleton + README structure reference
 - [`BerriAI/litellm`](https://github.com/BerriAI/litellm) — bidirectional protocol translation patterns
 - [`tauri-apps/tauri`](https://tauri.app/) — v2 + `cas://` architecture base
-- [`openai/codex`](https://github.com/openai/codex) — autocompact prompt base structure + compact protocol reverse-reference (Apache-2.0)
+- [`openai/codex`](https://github.com/openai/codex) — autocompact prompt base structure + compact protocol reverse-reference + verbatim apply_patch V4A chat-path system guidance (`codex-rs/core/prompt_with_apply_patch_instructions.md` L277-L351 @ commit `0b4f86095c8005d8f74e9c62b971d72c1670aa88`, Apache-2.0)
 - [`Piebald-AI/claude-code-system-prompts`](https://github.com/Piebald-AI/claude-code-system-prompts) — autocompact anchor bullets (All user messages verbatim + Next Step verbatim)
 - [`7as0nch/mimo2codex`](https://github.com/7as0nch/mimo2codex) — MiMo protocol reference
 - [`router-for-me/CLIProxyAPI`](https://github.com/router-for-me/CLIProxyAPI) — Gemini OAuth wire-level reference

diff --git a/README.md b/README.md
@@ -41,6 +41,7 @@ Codex App Transfer 是一个面向 **OpenAI Codex CLI** 的轻量桌面配置 +
 - 管理多套供应商,按 OpenAI 模型名(`gpt-5.5` / `gpt-5.4` / `gpt-5.4-mini` / `gpt-5.3-codex` / `gpt-5.2`)映射到供应商真实模型 ID
 - 把 Codex CLI 的 Responses API 流式 / 非流式请求转换为上游协议:Chat Completions、Gemini Native(`:streamGenerateContent`)、Gemini CLI OAuth(Cloud Code Assist)、Anthropic Messages(`/v1/messages`)、Grok Web(`/rest/app-chat/conversations/new`)、Responses 透传等
 - 多轮工具对话上下文 + `previous_response_id` 历史回放 + autocompact 展开 + thinking / reasoning_content 注入全部对齐 OpenAI Responses API 协议
+- Codex CLI 的 freeform `apply_patch` 工具(编辑文件 +/- diff UI)在 DeepSeek / Kimi / MiMo 等 chat-completions provider 上正常工作:adapter 双向桥接 Responses `custom_tool_call` ↔ chat `function_call` 形态,模型按 V4A 格式生成 patch,Codex CLI 渲染为 diff(issue #235)
 - 会话历史**两层持久化**:L1 内存 LRU + L2 sqlite 30 天 TTL(`~/.codex-app-transfer/sessions.db`),`.app` 重启不丢历史
 - Codex CLI 原配置守护:apply 前自动快照 `~/.codex/{config.toml,auth.json}`,退出 / 下次启动按 key 智能合并还原
 - 实时日志面板,2 秒自动刷新;统一 `tracing::warn!(error_id, detail)` + 稳定 token,operator 可 grep / 聚合
@@ -232,7 +233,7 @@ v2.1.12+ 的客户端 **强制** RSA-3072 PKCS#1-v1.5-SHA256 验签 `latest.json
 - [`lonr-6/cc-desktop-switch`](https://github.com/lonr-6/cc-desktop-switch) — v1.x 桌面壳骨架 + README 结构参考
 - [`BerriAI/litellm`](https://github.com/BerriAI/litellm) — 协议双向转换思路
 - [`tauri-apps/tauri`](https://tauri.app/) — v2 + `cas://` 架构基座
-- [`openai/codex`](https://github.com/openai/codex) — autocompact prompt 骨架 + compact 协议结构反查(Apache-2.0)
+- [`openai/codex`](https://github.com/openai/codex) — autocompact prompt 骨架 + compact 协议结构反查 + apply_patch V4A chat-path system guidance verbatim 引用(`codex-rs/core/prompt_with_apply_patch_instructions.md` L277-L351 @ commit `0b4f86095c8005d8f74e9c62b971d72c1670aa88`,Apache-2.0)
 - [`Piebald-AI/claude-code-system-prompts`](https://github.com/Piebald-AI/claude-code-system-prompts) — autocompact prompt 锚定 bullet(All user messages verbatim + Next Step verbatim)
 - [`7as0nch/mimo2codex`](https://github.com/7as0nch/mimo2codex) — MiMo 协议借鉴
 - [`router-for-me/CLIProxyAPI`](https://github.com/router-for-me/CLIProxyAPI) — Gemini OAuth wire 参考

diff --git a/crates/adapters/src/responses/apply_patch_v4a_reference.md b/crates/adapters/src/responses/apply_patch_v4a_reference.md
@@ -0,0 +1,79 @@
+> Adapter note (codex-app-transfer): in this chat-completions environment `apply_patch` is exposed as a function-call tool, NOT a shell command. Read every "`apply_patch` shell command" / "invoke apply_patch like `shell {"command":[...]}`" reference in the prose below as "the `apply_patch` function-call tool" — the patch envelope, hunk format, and V4A grammar are identical; only the invocation surface differs. Specifically: call the `apply_patch` function with a single `input` string containing the V4A patch (no `shell` wrapper, no JSON-quoting tricks); the trailing shell-style invocation example at the very end is for the upstream Codex CLI's shell-tool path and does not apply here.
+>
+> Source: verbatim mirror of openai/codex @ commit `0b4f86095c8005d8f74e9c62b971d72c1670aa88`, file `codex-rs/core/prompt_with_apply_patch_instructions.md`, lines L277-L351. Apache-2.0, Copyright 2025 OpenAI. Stored at `crates/adapters/src/responses/apply_patch_v4a_reference.md` in this repository. See `NOTICE` + `ACKNOWLEDGEMENTS.md` for full attribution.
+
+## `apply_patch`
+
+Use the `apply_patch` shell command to edit files.
+Your patch language is a stripped‑down, file‑oriented diff format designed to be easy to parse and safe to apply. You can think of it as a high‑level envelope:
+
+*** Begin Patch
+[ one or more file sections ]
+*** End Patch
+
+Within that envelope, you get a sequence of file operations.
+You MUST include a header to specify the action you are taking.
+Each operation starts with one of three headers:
+
+*** Add File: <path> - create a new file. Every following line is a + line (the initial contents).
+*** Delete File: <path> - remove an existing file. Nothing follows.
+*** Update File: <path> - patch an existing file in place (optionally with a rename).
+
+May be immediately followed by *** Move to: <new path> if you want to rename the file.
+Then one or more “hunks”, each introduced by @@ (optionally followed by a hunk header).
+Within a hunk each line starts with:
+
+For instructions on [context_before] and [context_after]:
+- By default, show 3 lines of code immediately above and 3 lines immediately below each change. If a change is within 3 lines of a previous change, do NOT duplicate the first change’s [context_after] lines in the second change’s [context_before] lines.
+- If 3 lines of context is insufficient to uniquely identify the snippet of code within the file, use the @@ operator to indicate the class or function to which the snippet belongs. For instance, we might have:
+@@ class BaseClass
+[3 lines of pre-context]
+- [old_code]
++ [new_code]
+[3 lines of post-context]
+
+- If a code block is repeated so many times in a class or function such that even a single `@@` statement and 3 lines of context cannot uniquely identify the snippet of code, you can use multiple `@@` statements to jump to the right context. For instance:
+
+@@ class BaseClass
+@@ 	 def method():
+[3 lines of pre-context]
+- [old_code]
++ [new_code]
+[3 lines of post-context]
+
+The full grammar definition is below:
+Patch := Begin { FileOp } End
+Begin := "*** Begin Patch" NEWLINE
+End := "*** End Patch" NEWLINE
+FileOp := AddFile | DeleteFile | UpdateFile
+AddFile := "*** Add File: " path NEWLINE { "+" line NEWLINE }
+DeleteFile := "*** Delete File: " path NEWLINE
+UpdateFile := "*** Update File: " path NEWLINE [ MoveTo ] { Hunk }
+MoveTo := "*** Move to: " newPath NEWLINE
+Hunk := "@@" [ header ] NEWLINE { HunkLine } [ "*** End of File" NEWLINE ]
+HunkLine := (" " | "-" | "+") text NEWLINE
+
+A full patch can combine several operations:
+
+*** Begin Patch
+*** Add File: hello.txt
++Hello world
+*** Update File: src/app.py
+*** Move to: src/main.py
+@@ def greet():
+-print("Hi")
++print("Hello, world!")
+*** Delete File: obsolete.txt
+*** End Patch
+
+It is important to remember:
+
+- You must include a header with your intended action (Add/Delete/Update)
+- You must prefix new lines with `+` even when creating a new file
+- File references can only be relative, NEVER ABSOLUTE.
+
+You can invoke apply_patch like:
+
+```
+shell {"command":["apply_patch","*** Begin Patch\n*** Add File: hello.txt\n+Hello, world!\n*** End Patch\n"]}
+```