scriptoriumadmin/ReMe
1
0
Fork 0
mirror of https://github.com/agentscope-ai/ReMe.git synced 2026-10-14 04:12:05 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 11

docs(integrations): make plugin installation the coding agent entry point

Browse source
This commit is contained in:
huangsen 2026-09-28 18:02:53 +08:00
parent 67bdbe4c55
commit 2ce7b25d94
8 changed files with 122 additions and 71 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
README.md
View file

@ -199,9 +199,9 @@ lifecycle according to the capabilities of each runtime.
| **DeepSeek Harness** | Install [`@agentscope-ai/reme-dsh-plugin`](https://reme.agentscope.io/en/integrations/dsh) with `dsh plugin --profile web add @agentscope-ai/reme-dsh-plugin`. | Configurable memory guidance, `reme_search`, automatic turn capture, scheduled Auto Dream, and ReMe Status. |
| **OpenClaw** | Install [`@agentscope-ai/reme-openclaw-plugin`](https://reme.agentscope.io/en/integrations/openclaw) with `openclaw plugins install clawhub:@agentscope-ai/reme-openclaw-plugin`. | Native memory tools, recall before user-triggered runs, and automatic turn capture. |
| **QwenPaw** | Embed ReMe in-process through its Python API. | Reuse the host lifecycle and model config while keeping memory local and file-based. |
| **Claude Code** | Install the repository-local [Claude Code plugin](integrations/claude_code/README.md). | Automatic recall, batched completed-turn recording, and MCP tools. |
| **Claude Code** | Start ReMe, then install the [Claude Code plugin](integrations/claude_code/README.md). | Automatic recall, batched completed-turn recording, and MCP tools. |
| **Hermes** | Install [the ReMe provider](https://reme.agentscope.io/en/integrations/hermes) and choose HTTP or embedded mode. | Recall before model calls and asynchronous `auto_memory` after each completed turn. |
| **Codex** | Install the repository-local [Codex plugin](integrations/codex/README.md). | Native hooks for automatic recall and batched recording, plus MCP tools. |
| **Codex** | Start ReMe, then install the [Codex plugin](integrations/codex/README.md). | Native hooks for automatic recall and batched recording, plus MCP tools. |
| **Other CLI agents** | Install or copy the [ReMe Memory skill](skills/reme_memory/SKILL.md). | Search, read, and write memory through the CLI. |
<p align="center"><b>Integration demos</b></p>

4
README_ZH.md
View file

@ -191,9 +191,9 @@ runtime 的能力，将记忆指引、召回和捕获接入 Agent 生命周期
| **DeepSeek Harness** | 使用 `dsh plugin --profile web add @agentscope-ai/reme-dsh-plugin` 安装 [`@agentscope-ai/reme-dsh-plugin`](https://reme.agentscope.io/zh/integrations/dsh)。 | 可配置记忆指引、`reme_search`、自动对话捕获、定时 Auto Dream 和 ReMe Status。 |
| **OpenClaw** | 使用 `openclaw plugins install clawhub:@agentscope-ai/reme-openclaw-plugin` 安装 [`@agentscope-ai/reme-openclaw-plugin`](https://reme.agentscope.io/zh/integrations/openclaw)。 | 原生记忆工具、用户触发运行前召回和自动对话捕获。 |
| **QwenPaw** | 通过 Python API 在进程内嵌入 ReMe。 | 复用宿主生命周期和模型配置，同时保持记忆本地、文件化。 |
| **Claude Code** | 安装仓库内的 [Claude Code 插件](integrations/claude_code/README_ZH.md)。 | 自动召回、完成轮次分批记录，以及 MCP 查询工具。 |
| **Claude Code** | 启动 ReMe 服务后，安装 [Claude Code 插件](integrations/claude_code/README_ZH.md)。 | 自动召回、完成轮次分批记录，以及 MCP 查询工具。 |
| **Hermes** | 安装 [ReMe provider](https://reme.agentscope.io/zh/integrations/hermes)，并选择 HTTP 或 Embedded 模式。 | 模型调用前召回，每轮对话完成后异步执行 `auto_memory`。 |
| **Codex** | 安装仓库内的 [Codex 插件](integrations/codex/README_ZH.md)。 | 原生 Hook 自动召回、分批记录，以及 MCP 查询工具。 |
| **Codex** | 启动 ReMe 服务后，安装 [Codex 插件](integrations/codex/README_ZH.md)。 | 原生 Hook 自动召回、分批记录，以及 MCP 查询工具。 |
| **其他 CLI Agent** | 安装或复制 [ReMe Memory skill](skills/reme_memory/SKILL.md)。 | 通过 CLI 搜索、读取和写入记忆。 |
<p align="center"><b>集成演示</b></p>

13
docs/en/integrations.md
View file

@ -16,9 +16,9 @@ ReMe keeps memory in an independent service and a user-owned workspace. Multiple
| Tool-protocol host | MCP |
| DeepSeek Harness | [`@agentscope-ai/reme-dsh-plugin`](./integrations/dsh.md) profile bundle |
| OpenClaw | [`@agentscope-ai/reme-openclaw-plugin`](./integrations/openclaw.md) |
| Claude Code | [Repository-local plugin](./integrations/claude-code.md) |
| Claude Code | [Start ReMe, then install the plugin](./integrations/claude-code.md) |
| Hermes Agent | Memory provider adapter |
| Codex | [Repository-local plugin](./integrations/codex.md) |
| Codex | [Start ReMe, then install the plugin](./integrations/codex.md) |
| Other coding agents | `reme_memory` Skill or MCP |
## General memory loop
@ -78,11 +78,16 @@ ReMe HTTP boundary and release lifecycle.
## Claude Code
[Claude Code](./integrations/claude-code.md) has its own repository-local plugin with automatic pre-prompt recall, batched completed-turn recording through `auto_memory`, MCP tools, and a `reme-memory` skill.
Start the ReMe HTTP service, then install the [ReMe plugin in Claude Code](./integrations/claude-code.md).
The installed plugin supplies automatic recall, completed-turn recording, and memory queries through its
bundled hooks, MCP tools, and skill. This host uses the plugin installation flow.
## Codex
[Codex](./integrations/codex.md) has a separate repository-local marketplace and plugin. Native hooks provide automatic recall and recording after review in `/hooks`; MCP and the skill also support explicit queries. Both integrations keep retry state in their own host directories and leave consolidation to the ReMe service.
Start the ReMe HTTP service, install the [ReMe plugin in Codex](./integrations/codex.md), and trust its
hooks in `/hooks`. Open a new conversation to use automatic recall, recording, and the plugin's memory tools.
This host also uses the plugin installation flow. Each host keeps its own plugin configuration and retry
queue; the ReMe service owns memory storage and consolidation.
## Hermes Agent

12
docs/zh/integrations.md
View file

@ -16,9 +16,9 @@ ReMe 把记忆能力放在独立服务和用户拥有的 workspace 中。Agent
| 支持工具协议的 Agent | MCP |
| DeepSeek Harness | [`@agentscope-ai/reme-dsh-plugin`](./integrations/dsh.md) profile bundle |
| OpenClaw | [`@agentscope-ai/reme-openclaw-plugin`](./integrations/openclaw.md) |
| Claude Code | [仓库内独立插件](./integrations/claude-code.md) |
| Claude Code | [启动 ReMe 后安装插件](./integrations/claude-code.md) |
| Hermes Agent | Memory provider adapter |
| Codex | [仓库内独立插件](./integrations/codex.md) |
| Codex | [启动 ReMe 后安装插件](./integrations/codex.md) |
| 其他 coding agent | `reme_memory` Skill 或 MCP |
## 通用接入循环
@ -88,11 +88,15 @@ openclaw plugins install clawhub:@agentscope-ai/reme-openclaw-plugin
## Claude Code
[Claude Code 插件](./integrations/claude-code.md) 独立位于 `integrations/claude_code/`，提供消息提交前自动召回、通过 `auto_memory` 分批记录完成轮次、MCP 工具和 `reme-memory` Skill。
先启动 ReMe HTTP 服务，再[安装 Claude Code 的 ReMe 插件](./integrations/claude-code.md)。
插件统一提供自动召回、完成轮次记录和记忆查询，所需 Hook、MCP 工具与 Skill 随插件安装。
本宿主使用插件安装流程接入。
## Codex
[Codex 插件](./integrations/codex.md) 独立位于 `integrations/codex/`，附带本地 marketplace。原生 Hook 在 `/hooks` 审核后提供自动召回与记录；MCP 和 Skill 支持显式查询。两端各自保存配置与重试队列，记忆整理仍由 ReMe 服务负责。
先启动 ReMe HTTP 服务，再[安装 Codex 的 ReMe 插件](./integrations/codex.md)，在 `/hooks` 中信任插件 Hook。
新建会话后即可使用自动召回、记录和插件提供的记忆工具。本宿主同样使用插件安装流程接入。
两端各自保存插件配置与重试队列，记忆存储和整理由 ReMe 服务负责。
## Hermes Agent

42
integrations/claude_code/README.md
View file

@ -2,14 +2,15 @@
[中文说明](./README_ZH.md)
This repository-local plugin gives Claude Code automatic recall before user prompts, background
recording of completed turns, and explicit ReMe MCP tools. Durable memory stays in a user-owned
ReMe workspace. Its plugin files and installation guide are maintained under `integrations/claude_code/` and can be
installed directly from this checkout.
Start a ReMe service, then install the ReMe plugin in Claude Code. This is the supported integration
for this host. The plugin bundles automatic recall, completed-turn recording, MCP tools, and the
`reme-memory` skill in one installation. ReMe manages durable memory and consolidation in your workspace.
The plugin and its marketplace are maintained in `integrations/claude_code/` and installed from
this checkout. Keep the ReMe service running while using the plugin.
```text
UserPromptSubmit → search → untrusted memory context → agent reply
Stop → completed user/assistant text → local retry queue → auto_memory → daily Markdown
Start ReMe → Install the ReMe plugin → Use memory in Claude Code
```
## Requirements
@ -19,7 +20,7 @@ Stop → completed user/assistant text → local retry queue → auto_memory →
- A running ReMe HTTP service exposing `search`, `auto_memory`, and `health_check`, with MCP enabled.
- A working model configuration on the ReMe server for memory extraction. Default BM25 recall does not need embeddings.
## Start ReMe
## 1. Start ReMe
```bash
pip install "reme-ai[core]"
@ -30,7 +31,7 @@ The default endpoint is `http://127.0.0.1:2333`, with MCP at `/mcp`. Keep the un
service on loopback or behind a protected proxy. Different workspaces isolate different memory
scopes; sharing one workspace between hosts is intentional cross-agent recall.
## Install the plugin
## 2. Install the plugin
In Claude Code, using an absolute path to this checkout:
@ -39,13 +40,23 @@ In Claude Code, using an absolute path to this checkout:
/plugin install reme@reme-marketplace
```
For a single development session, use `claude --plugin-dir /absolute/path/to/ReMe/integrations/claude_code/reme`.
Restart or reload the plugin after source changes. Open `/mcp` to verify the `reme` connection.
Restart Claude Code or reload the installed plugin, then open a new conversation. The installation
loads the bundled hooks, MCP connection, and skill together. Reinstall or reload the plugin after updates.
## 3. Use memory in Claude Code
In a new conversation, ask “Check whether ReMe is available.” The installed plugin's tools check
the running service; `/mcp` shows the plugin's connection status. Then use Claude Code normally:
the plugin recalls relevant memory before prompts and records completed turns in batches of five.
For an explicit query, ask “What did we decide about Project Juniper? Cite the memory sources.”
Automatic recording sends completed user/assistant text to the configured ReMe service.
Set `auto_memory` to `false` for recall-only use.
## Configuration
## Optional configuration
The default service address works immediately after plugin installation. Create a configuration
file only when you want to change the defaults below.
Optional settings live at `${CLAUDE_CONFIG_DIR:-$HOME/.claude}/reme/config.json`. Start from the bundled example:
@ -97,15 +108,16 @@ There is no separate `REME_HOST`/`REME_PORT` hook override.
## Verify and troubleshoot
1. Call `reme health_check` against the configured service and check `/mcp` in Claude Code.
1. In Claude Code, ask the installed ReMe plugin to check the connection; confirm its status in `/mcp`.
2. For a quick test, set `memory_interval` to `1`, then start a new conversation and ask the agent to
remember a synthetic fact, such as “Project Juniper reviews are on Thursday.”
3. Wait for the write and inspect `reme/hooks.log` for `memory_saved` and the ReMe workspace's `daily/`.
4. Start a separate conversation and ask for that fact. Confirm recall is from ReMe, with source paths.
The `reme-memory` skill supports explicit search/read and health checks. Missing MCP tools can also
mean a disabled plugin or restricted `service.jobs`; it does not prove the server is stopped.
If hooks are unavailable in the installed host version, MCP and the skill remain the manual fallback.
The installed plugin provides both automatic memory and explicit search/read and health checks.
If its tools are missing, check that the plugin is installed and enabled, the service is running,
and `service.jobs` exposes the required jobs. If its hooks do not load, use a host version supporting
the plugin requirements and reinstall or reload the plugin.
Capture currently reads Claude JSONL user/assistant records with UUIDs. The hook requires
`transcript_path` and `last_assistant_message`. Missing or unsupported transcripts are skipped rather

36
integrations/claude_code/README_ZH.md
View file

@ -2,13 +2,14 @@
[English](./README.md)
本插件在用户提交消息前自动召回记忆，在完成对话后后台分批记录，同时提供显式 MCP 查询工具。
长期记忆保存在用户拥有的 ReMe workspace。插件与安装文档独立维护在
`integrations/claude_code/`，可直接从仓库安装。
先启动 ReMe 服务，再在 Claude Code 中安装 ReMe 插件，即可使用长期记忆。这是本宿主支持的集成方式。
插件统一安装自动召回、完成轮次记录、MCP 工具和内置 `reme-memory` Skill；ReMe 服务负责工作区中的记忆存储与整理。
插件与本地 marketplace 维护在 `integrations/claude_code/`，从仓库安装。
使用插件期间，保持 ReMe 服务运行。
```text
UserPromptSubmit → search → 不可信历史上下文 → 助手回答
Stop → 完成的用户/助手文本 → 本地重试队列 → auto_memory → daily Markdown
启动 ReMe 服务 → 安装 ReMe 插件 → 在 Claude Code 中使用记忆
```
## 环境要求
@ -18,7 +19,7 @@ Stop → 完成的用户/助手文本 → 本地重试队列 → auto_memory →
- 已启动 ReMe HTTP 服务，开放 `search`、`auto_memory`、`health_check`，并启用 MCP。
- ReMe 服务端有可用于自动记忆提取的模型配置。默认 BM25 召回不要求 Embedding。
## 启动 ReMe
## 1. 启动 ReMe 服务
```bash
pip install "reme-ai[core]"
@ -28,7 +29,7 @@ reme start workspace_dir=/absolute/path/to/workspace service.backend=http
默认地址为 `http://127.0.0.1:2333`，MCP 路径为 `/mcp`。HTTP 服务没有内建 API Key 鉴权，
应保持 loopback 监听或部署受保护的代理。不同 workspace 隔离记忆；不同宿主使用同一 workspace 时会共享召回。
## 安装插件
## 2. 安装 ReMe 插件
在 Claude Code 中使用仓库的绝对路径：
@ -37,12 +38,20 @@ reme start workspace_dir=/absolute/path/to/workspace service.backend=http
/plugin install reme@reme-marketplace
```
单次开发验证可使用 `claude --plugin-dir /absolute/path/to/ReMe/integrations/claude_code/reme`。
源码修改后重新加载或安装插件，通过 `/mcp` 确认 `reme` 连接。
重启 Claude Code 或重新加载已安装的插件，然后新建会话。插件安装会一并加载 Hook、MCP 连接和内置 Skill。
更新后重新安装或加载插件。
## 3. 在 Claude Code 中使用记忆
在新会话中询问“检查 ReMe 是否可用”，由已安装插件的工具检查服务连接；也可通过 `/mcp` 查看插件连接状态。
之后正常使用 Claude Code：插件在消息提交前自动召回相关记忆，每完成五轮对话分批记录一次。
需要主动查询时，可以询问“之前关于 Juniper 项目做了什么决定？请附上记忆来源”。
自动记录会把完成的用户/助手文本发送到配置的 ReMe 服务；仅需召回时，将 `auto_memory` 设为 `false`。
## 配置
## 可选配置
使用默认服务地址时，安装插件后即可使用。需要调整下列默认行为时，再创建配置文件。
可选配置位于 `${CLAUDE_CONFIG_DIR:-$HOME/.claude}/reme/config.json`。从仓库根目录复制示例：
@ -81,13 +90,14 @@ cp integrations/claude_code/reme/config.example.json "${CLAUDE_CONFIG_DIR:-$HOME
## 验证与排查
1. 对配置的服务运行 `reme health_check`，通过 Claude Code `/mcp` 确认连接。
1. 在 Claude Code 中让已安装的 ReMe 插件检查连接，通过 `/mcp` 确认插件连接状态。
2. 快速验证时把 `memory_interval` 设为 `1`，新建会话，要求记住一条合成事实，例如“Juniper 项目每周四评审”。
3. 等待后台写入，检查 `reme/hooks.log` 的 `memory_saved`，以及 ReMe workspace 的 `daily/` 笔记。
4. 再新建一个独立会话，询问该事实，并确认回答附有 ReMe 来源路径。
`reme-memory` Skill 支持显式检索、阅读和健康检查。MCP 工具缺失也可能源于插件未启用或
`service.jobs` 限制，不能直接判断服务未启动。旧宿主不支持 Hook 时，可继续使用 MCP 和 Skill 手动查询。
自动记忆和显式检索、阅读、健康检查均由已安装的插件提供。工具缺失时，检查插件是否已安装并启用、
ReMe 服务是否运行，以及 `service.jobs` 是否开放所需 Job。Hook 无法加载时，请使用满足插件要求的宿主版本，
再重新安装或加载插件。
记录要求 Hook 提供 `transcript_path` 和 `last_assistant_message`，读取标准用户/助手 JSONL，
不读取私有推理；未知格式或缺失文件会跳过并记录状态，不猜测对话。

45
integrations/codex/README.md
View file

@ -2,14 +2,15 @@
[中文说明](./README_ZH.md)
This repository-local plugin gives Codex automatic recall before user prompts, automatic
recording of completed turns, and explicit ReMe MCP tools. Durable memory stays in a user-owned
ReMe workspace. Its plugin files and installation guide are maintained under `integrations/codex/` and can be
installed directly from this checkout.
Start a ReMe service, then install the ReMe plugin in Codex. This is the supported integration
for this host. The plugin bundles automatic recall, completed-turn recording, MCP tools, and the
`reme-memory` skill in one installation. ReMe manages durable memory and consolidation in your workspace.
The plugin and its marketplace are maintained in `integrations/codex/` and installed from
this checkout. Keep the ReMe service running while using the plugin.
```text
UserPromptSubmit → search → untrusted memory context → agent reply
Stop → completed user/assistant text → local retry queue → auto_memory → daily Markdown
Start ReMe → Install the ReMe plugin → Use memory in Codex
```
## Requirements
@ -19,7 +20,7 @@ Stop → completed user/assistant text → local retry queue → auto_memory →
- A running ReMe HTTP service exposing `search`, `auto_memory`, and `health_check`, with MCP enabled.
- A working model configuration on the ReMe server for memory extraction. Default BM25 recall does not need embeddings.
## Start ReMe
## 1. Start ReMe
```bash
pip install "reme-ai[core]"
@ -30,7 +31,7 @@ The default endpoint is `http://127.0.0.1:2333`, with MCP at `/mcp`. Keep the un
service on loopback or behind a protected proxy. Different workspaces isolate different memory
scopes; sharing one workspace between hosts is intentional cross-agent recall.
## Install the plugin
## 2. Install the plugin
Register this repository-local marketplace, then install its plugin:
@ -39,15 +40,24 @@ codex plugin marketplace add /absolute/path/to/ReMe/integrations/codex
codex plugin add reme@reme-codex
```
In the Codex app, the same local marketplace can be selected in the plugin directory. Start a new
thread after installing. Open `/hooks` in the CLI to review and trust the bundled hooks; enabling
a plugin alone does not trust its hooks. Open `/mcp` to verify ReMe. For updates, refresh the local
marketplace and reinstall the plugin, then review any changed hook definitions.
Start Codex and review and trust the installed plugin's hooks in `/hooks`, then open a new conversation.
The installation loads the bundled hooks, MCP connection, and skill together. After plugin updates,
reinstall it and review any changed hook definitions.
## 3. Use memory in Codex
In a new conversation, ask “Check whether ReMe is available.” The installed plugin's tools check
the running service; `/mcp` shows the plugin's connection status. Then use Codex normally:
the plugin recalls relevant memory before prompts and records completed turns in batches of five.
For an explicit query, ask “What did we decide about Project Juniper? Cite the memory sources.”
Automatic recording sends completed user/assistant text to the configured ReMe service.
Set `auto_memory` to `false` for recall-only use.
## Configuration
## Optional configuration
The default service address works immediately after plugin installation. Create a configuration
file only when you want to change the defaults below.
Optional settings live at `${CODEX_HOME:-$HOME/.codex}/reme/config.json`. Start from the bundled example:
@ -102,15 +112,16 @@ There is no separate `REME_HOST`/`REME_PORT` hook override.
## Verify and troubleshoot
1. Call `reme health_check` against the configured service and check `/mcp` in Codex.
1. In Codex, ask the installed ReMe plugin to check the connection; confirm its status in `/mcp`.
2. For a quick test, set `memory_interval` to `1`, then start a new conversation and ask the agent to
remember a synthetic fact, such as “Project Juniper reviews are on Thursday.”
3. Wait for the write and inspect `reme/hooks.log` for `memory_saved` and the ReMe workspace's `daily/`.
4. Start a separate conversation and ask for that fact. Confirm recall is from ReMe, with source paths.
The `reme-memory` skill supports explicit search/read and health checks. Missing MCP tools can also
mean a disabled plugin or restricted `service.jobs`; it does not prove the server is stopped.
If hooks are unavailable in the installed host version, MCP and the skill remain the manual fallback.
The installed plugin provides both automatic memory and explicit search/read and health checks.
If its tools are missing, check that the plugin is installed and enabled, the service is running,
and `service.jobs` exposes the required jobs. If its hooks do not load, use a host version supporting
the plugin requirements and reinstall or reload the plugin.
Capture currently reads Codex rollout JSONL `event_msg` records (`user_message` and final `agent_message`). The hook requires
`transcript_path` and `last_assistant_message`. Missing or unsupported transcripts are skipped rather

37
integrations/codex/README_ZH.md
View file

@ -2,13 +2,14 @@
[English](./README.md)
本插件在用户提交消息前自动召回记忆，在完成对话后分批记录，同时提供显式 MCP 查询工具。
长期记忆保存在用户拥有的 ReMe workspace。插件与安装文档独立维护在
`integrations/codex/`，可直接从仓库安装。
先启动 ReMe 服务，再在 Codex 中安装 ReMe 插件，即可使用长期记忆。这是本宿主支持的集成方式。
插件统一安装自动召回、完成轮次记录、MCP 工具和内置 `reme-memory` Skill；ReMe 服务负责工作区中的记忆存储与整理。
插件与本地 marketplace 维护在 `integrations/codex/`，从仓库安装。
使用插件期间，保持 ReMe 服务运行。
```text
UserPromptSubmit → search → 不可信历史上下文 → 助手回答
Stop → 完成的用户/助手文本 → 本地重试队列 → auto_memory → daily Markdown
启动 ReMe 服务 → 安装 ReMe 插件 → 在 Codex 中使用记忆
```
## 环境要求
@ -18,7 +19,7 @@ Stop → 完成的用户/助手文本 → 本地重试队列 → auto_memory →
- 已启动 ReMe HTTP 服务，开放 `search`、`auto_memory`、`health_check`，并启用 MCP。
- ReMe 服务端有可用于自动记忆提取的模型配置。默认 BM25 召回不要求 Embedding。
## 启动 ReMe
## 1. 启动 ReMe 服务
```bash
pip install "reme-ai[core]"
@ -28,7 +29,7 @@ reme start workspace_dir=/absolute/path/to/workspace service.backend=http
默认地址为 `http://127.0.0.1:2333`，MCP 路径为 `/mcp`。HTTP 服务没有内建 API Key 鉴权，
应保持 loopback 监听或部署受保护的代理。不同 workspace 隔离记忆；不同宿主使用同一 workspace 时会共享召回。
## 安装插件
## 2. 安装 ReMe 插件
添加仓库内的本地 marketplace，再安装插件：
@ -37,13 +38,20 @@ codex plugin marketplace add /absolute/path/to/ReMe/integrations/codex
codex plugin add reme@reme-codex
```
Codex App 也可在插件目录中选择该本地 marketplace。安装后新建线程。
在 CLI 的 `/hooks` 中检查并信任插件 Hook；仅启用插件不会自动信任 Hook。通过 `/mcp` 确认 ReMe 连接。
修改源码后刷新本地 marketplace 并重新安装，新建线程，重新审核有变化的 Hook。
启动 Codex，在 `/hooks` 中检查并信任已安装插件的 Hook，然后新建会话。
插件安装会一并加载 Hook、MCP 连接和内置 Skill。更新后重新安装插件，并审核有变化的 Hook。
## 3. 在 Codex 中使用记忆
在新会话中询问“检查 ReMe 是否可用”，由已安装插件的工具检查服务连接；也可通过 `/mcp` 查看插件连接状态。
之后正常使用 Codex：插件在消息提交前自动召回相关记忆，每完成五轮对话分批记录一次。
需要主动查询时，可以询问“之前关于 Juniper 项目做了什么决定？请附上记忆来源”。
自动记录会把完成的用户/助手文本发送到配置的 ReMe 服务；仅需召回时，将 `auto_memory` 设为 `false`。
## 配置
## 可选配置
使用默认服务地址时，安装插件后即可使用。需要调整下列默认行为时，再创建配置文件。
可选配置位于 `${CODEX_HOME:-$HOME/.codex}/reme/config.json`。从仓库根目录复制示例：
@ -84,13 +92,14 @@ Codex 0.145.0 会跳过设置了 `async` 的 hook，因此本适配器使用同
## 验证与排查
1. 对配置的服务运行 `reme health_check`，通过 Codex `/mcp` 确认连接。
1. 在 Codex 中让已安装的 ReMe 插件检查连接，通过 `/mcp` 确认插件连接状态。
2. 快速验证时把 `memory_interval` 设为 `1`，新建会话，要求记住一条合成事实，例如“Juniper 项目每周四评审”。
3. 等待写入，检查 `reme/hooks.log` 的 `memory_saved`，以及 ReMe workspace 的 `daily/` 笔记。
4. 再新建一个独立会话，询问该事实，并确认回答附有 ReMe 来源路径。
`reme-memory` Skill 支持显式检索、阅读和健康检查。MCP 工具缺失也可能源于插件未启用或
`service.jobs` 限制，不能直接判断服务未启动。旧宿主不支持 Hook 时，可继续使用 MCP 和 Skill 手动查询。
自动记忆和显式检索、阅读、健康检查均由已安装的插件提供。工具缺失时，检查插件是否已安装并启用、
ReMe 服务是否运行，以及 `service.jobs` 是否开放所需 Job。Hook 无法加载时，请使用满足插件要求的宿主版本，
再重新安装或加载插件。
记录要求 Hook 提供 `transcript_path` 和 `last_assistant_message`，读取 rollout 的 `event_msg`
（`user_message` 与最终 `agent_message`），不读取私有推理。