Compare commits

...

1 Commits

Author SHA1 Message Date
Dizhan Xue
62e2266f2a docs: explain codex desktop history sharing
Document how the existing Codex sub-agent shares native conversation history
with the desktop app, including Codex home and project directory requirements.

Co-authored-by: Codex <noreply@openai.com>
2026-10-19 10:36:33 +08:00
2 changed files with 55 additions and 0 deletions

View File

@ -222,6 +222,37 @@ Distinguish three claims when assessing compatibility:
Hand-written integrations need their own checks at all three applicable levels.
### Codex desktop history { #codex-desktop-history }
Use the existing `codex` sub-agent for Codex CLI and the Codex desktop App.
Its ACP adapter starts Codex App Server and creates native Codex threads. The
adapter includes a compatible Codex executable; the preset needs Node.js and
`npx`, but does not require a separate `codex` installation on `PATH`.
To view a Raven-delegated Codex conversation in the desktop App:
1. Run Raven and the App as the same OS user, with the same `CODEX_HOME`
(normally `~/.codex`). If Raven uses a different environment, set
`env.CODEX_HOME` on its existing Codex agent entry to the App's absolute
Codex home path. Preserve the entry's other environment settings.
2. Open the same project directory in the App as the working directory of the
Raven task. The App filters history by project; an agent entry's launch
`cwd` does not override the task's working directory.
3. Let the Raven turn finish before opening and continuing that conversation
in the App. It contains that Codex sub-agent's conversation, rather than
Raven's entire orchestration transcript.
A different `CODEX_HOME`, OS user, container, WSL environment, or remote machine
can have a separate history store. Using the same account alone does not make
those local stores the same. If a thread is missing, check the home directory,
project directory, and Codex versions first.
History sharing does not attach Raven to the desktop window's running task or
coordinate simultaneous turns from both clients. Raven continues its existing
sub-agent instances through the same ACP session-loading path. See OpenAI's
[App Server documentation](https://learn.chatgpt.com/docs/app-server) for thread
history and resume behavior.
## Connect and verify
1. Install and authenticate the external agent according to its own version.

View File

@ -189,6 +189,30 @@ elicitation 发出,由你或 Raven 的提问自动作答回答。不支持表
手写集成需要对所有适用层级分别检查。
### Codex 桌面历史记录 { #codex-desktop-history }
Codex CLI 和 Codex 桌面 App 使用同一个现有 `codex` 子 Agent。其 ACP 适配器启动
Codex App Server,创建原生 Codex 会话。适配器自带兼容的 Codex 可执行文件;该预设
需要 Node.js 和 `npx`,无需另行在 `PATH` 中安装 `codex`。
在桌面 App 中查看 Raven 委派给 Codex 的对话:
1. Raven 和 App 使用同一个系统用户、同一个 `CODEX_HOME`(通常是 `~/.codex`)。
若 Raven 的运行环境不同,在已有 Codex Agent 条目的 `env.CODEX_HOME` 中填写
App 所用 Codex 主目录的绝对路径,并保留条目中的其他环境设置。
2. 在 App 中打开与 Raven 任务工作目录相同的项目目录。App 会按项目筛选历史;
Agent 条目的启动 `cwd` 不会覆盖任务本身的工作目录。
3. 等 Raven 这一轮结束后,再到 App 中打开并继续该对话。这里保存的是该 Codex
子 Agent 的对话,而不是 Raven 全部编排过程的记录。
不同的 `CODEX_HOME`、系统用户、容器、WSL 环境或远程机器可能使用独立的历史存储。
仅登录同一个账号不会使这些本地存储自动成为同一份。找不到会话时,先检查主目录、
项目目录和 Codex 版本。
历史互通不会让 Raven 接管桌面窗口中正在执行的任务,也不会协调两端同时发起的
执行轮次。Raven 仍通过现有 ACP 会话加载流程继续自己的子 Agent instance。
会话历史和恢复行为见 OpenAI 的 [App Server 文档](https://learn.chatgpt.com/docs/app-server)。
## 接入与验证 { #connect-and-verify }
1. 按外部 Agent 对应版本安装并认证,确保其可执行文件在常驻 Raven 进程的 `PATH` 中。