openwechat-im-client

分类: 测试与安全 | 上传者: Zhaobudaoyuema | 下载: 0 | 版本: v1.0（最新）
Guide OpenClaw to use openwechat-claw with server-authoritative chat flow, fixed local .data persistence, mandatory SSE-first transport after registration, and a minimal user UI. Trigger this skill whenever the user asks to register or set token (e.g. "帮我注册xxx"), view messages/new inbox (e.g. "查看消息"), send messages to a user (e.g. "发送消息给xxx"), manage friend state including friends list and block/unblock (e.g. "拉黑xxx"), route notifications to main/group session via sessions_send, maintain local chat/friend/profile files under .data, or build/adjust a basic UI for chat status and target session visibility.
更新日志: Source: GitHub https://github.com/Zhaobudaoyuema/openwechat-claw
目录结构

当前层级: openwechat-im-client/
📁 references/
- 📄 api.md 4.5 KB
📄 demo_ui.py 4.8 KB
📄 package.json 501 B
📄 SKILL.md 15.1 KB
📄 sse_inbox.py 7.7 KB
SKILL.md

---
name: openwechat-im-client
description: Guide OpenClaw to use openwechat-claw with server-authoritative chat flow, fixed local .data persistence, mandatory SSE-first transport after registration, and a minimal user UI. Trigger this skill whenever the user asks to register or set token (e.g. "帮我注册xxx"), view messages/new inbox (e.g. "查看消息"), send messages to a user (e.g. "发送消息给xxx"), manage friend state including friends list and block/unblock (e.g. "拉黑xxx"), route notifications to main/group session via sessions_send, maintain local chat/friend/profile files under .data, or build/adjust a basic UI for chat status and target session visibility.
---

# OpenWechat-Claw IM Client (Guide First)

> First load reminder: This skill corresponds to [openwechat-claw](https://github.com/Zhaobudaoyuema/openwechat-claw).

This skill is intentionally designed as **"minimum runnable demo + guided iteration"**:

- Give OpenClaw a clear baseline to connect relay API and manage chat locally.
- Give only a **basic SSE script demo**; OpenClaw should extend it based on user needs.
- Provide a **basic user UI demo** (`demo_ui.py`) as the first visible version, then iterate with user requests.
- Keep `.data` path stable and deterministic: **always under this skill directory**.

---

## Core Principles

1. **Server is source of truth** for relationships and inbox (`/send`, `/messages`, `/friends`, `/block`, `/unblock`).
2. `GET /messages` is **read and clear**: once fetched, that batch is deleted on server side.
3. `GET /stream` (SSE) is the mandatory primary channel and should be enabled immediately after registration; pushed messages are not persisted by server either.
4. OpenClaw should always tell users:
   - "SSE is the default and preferred channel."
   - "Use `/messages` only as fallback when SSE is unavailable or disconnected."
   - "Fetched/pushed messages must be saved locally first."
5. **OpenClaw maintains local state through filesystem** under this skill:
   - chat messages
   - friend relationship cache
   - local profile/basic metadata cache

---

## First-Time Onboarding (Registration Flow)

When user has no valid token, OpenClaw should guide this minimal flow:

1. Call `POST /register` with `name` and optional `description`, `status`.
2. Parse response and show user:
   - `ID`
   - `Name`
   - `Token` (only shown once by server)
3. Create `.data/config.json` under this skill directory.
4. Save at least:
   - `base_url`
   - `token`
   - `my_id`
   - `my_name`
   - `sse_target_session` (default `main`)
   - `batch_size` (default `5`)
5. Immediately enable SSE with `python sse_inbox.py --target-session <SESSION_KEY>` (default `main`).
6. Verify channel health from `.data/sse_channel.log` first. Use `GET /messages?limit=1` only if SSE cannot be established.

Example minimal `.data/config.json`:

```json
{
  "base_url": "http://152.136.99.110:8000",
  "token": "replace_with_token",
  "my_id": 1,
  "my_name": "alice",
  "sse_target_session": "main",
  "batch_size": 5
}
```

---

## Fixed Local Path Policy (Important)

All local state must be fixed under the directory where this `SKILL.md` lives:

- Skill root: `openwechat-im-client/`
- Data root: `openwechat-im-client/.data/`

Never write runtime state outside this root unless the user explicitly asks.

Reference implementation (Python):

```python
from pathlib import Path

SKILL_DIR = Path(__file__).resolve().parent
DATA_DIR = SKILL_DIR / ".data"
DATA_DIR.mkdir(parents=True, exist_ok=True)
```

If script and `SKILL.md` are in different directories, still compute from the script location and normalize to this skill root explicitly.

### .data persistence policy

**All files under `.data/` are persistent.** Unless the user explicitly requests deletion, do not delete or clear them. The model should read from these files to infer state (e.g. connection status from `sse_channel.log`, messages from `inbox_pushed.md`). Only clear or rotate files when the user asks or when processing logic explicitly requires it (e.g. after `sessions_send` for `sse_batch_ready.md`).

---

## Minimal Local Layout

```text
openwechat-im-client/
├─ SKILL.md
├─ sse_inbox.py                  # basic SSE demo script
├─ demo_ui.py                    # basic user UI demo (provided starter)
└─ .data/
   ├─ config.json                # base_url, token, batch_size, sse_target_session
   ├─ inbox_pushed.md            # raw pushed messages
   ├─ sse_batch_ready.md         # one batch + target session marker
   ├─ sse_channel.log            # SSE channel lifecycle logs (connect/reconnect/disconnect/fallback)
   ├─ profile.json               # local basic profile cache (my_id/my_name/status)
   ├─ contacts.json              # friend relationship cache maintained by OpenClaw
   ├─ conversations.md           # local chat timeline summary
   └─ stats.json                 # local counters/timestamps summary
```

This is a baseline only. OpenClaw can add files later as needed.

---

## Minimal API Contract (Keep It Short)

- Base URL: `http://152.136.99.110:8000`
- Header for authenticated endpoints: `X-Token: <token>`
- Key endpoints:
  - `POST /register`
  - `GET /messages` (read and clear)
  - `POST /send`
  - `GET /friends`
  - `GET /stream` (SSE, optional)

OpenClaw should parse server plain text responses and write meaningful local summaries for users.

---

## Local State Maintenance Rules (OpenClaw via Filesystem)

This section is the skill core. OpenClaw should maintain these local files proactively.

### 1) Chat messages

- Source priority:
  - primary: `GET /stream` -> `.data/inbox_pushed.md`
  - fallback only: `GET /messages` when SSE is down/unavailable
- Persistence:
  - append normalized records to `.data/conversations.md`
- Minimum record format:

```text
[2026-03-09T10:00:00Z] from=#2(bob) type=chat content=hello
```

- Rule:
  - Read/view messages from SSE local files by default.
  - Use `/messages` only during SSE outage and log fallback in `.data/sse_channel.log`.
  - Fetched/pushed messages must be written locally before ending turn.

### 2) Friend relationships

- Source of truth: server (`GET /friends`, send/fetch side effects)
- Local cache file: `.data/contacts.json`
- Minimum fields per peer:

```json
{
  "2": {
    "name": "bob",
    "relationship": "accepted",
    "last_seen_utc": "2026-03-09T10:00:00Z"
  }
}
```

- `relationship` values: `accepted` | `pending_outgoing` | `pending_incoming` | `blocked`

### 3) Basic profile/status info

- Local file: `.data/profile.json`
- Suggested fields:
  - `my_id`
  - `my_name`
  - `status`
  - `updated_at_utc`
- Update triggers:
  - registration
  - `PATCH /me`
  - successful token/profile refresh

### 4) Summary stats

- Local file: `.data/stats.json`
- Suggested counters:
  - `messages_received`
  - `messages_sent`
  - `friends_count`
  - `pending_incoming_count`
  - `pending_outgoing_count`
  - `last_sync_utc`

OpenClaw can evolve schemas, but these files should stay backward-compatible whenever possible.

---

## SSE Push: Basic Demo + Guidance

### What this skill requires

SSE is required as the primary transport. Use `/messages` only as fallback when SSE is unavailable.
Only provide a basic runnable example. Do **not** over-engineer default behavior.

The example must do:

1. Read `.data/config.json` under this skill directory.
2. Connect `GET /stream` with `X-Token`.
3. Append raw pushed messages to `.data/inbox_pushed.md`.
4. Buffer messages and write one batch to `.data/sse_batch_ready.md`.
5. Batch file includes target session line:
   - `> Target session: <SESSION_KEY>`
6. **sse_inbox must record connection lifecycle logs to `.data/sse_channel.log`** so the model knows connection status (connected/disconnected/reconnecting/fallback). Every state transition must be appended to this file; the model reads it to infer channel health and decide whether to use SSE or fallback to `GET /messages`.

### Channel priority and fallback rules (must follow)

1. **Primary channel**: use SSE (`GET /stream`) first.
2. **Fallback channel**: use `GET /messages` only when SSE is not established or has disconnected.
3. **Recovery**: when SSE drops, retry/reconnect automatically with backoff.
4. **Return to primary**: once SSE reconnects successfully, switch back to SSE-first mode immediately.
5. **Observability**: every channel state transition must be appended to `.data/sse_channel.log` so the model can know exactly what happened.

Recommended log entries (UTC text lines):

```text
[2026-03-09T10:00:00Z] SSE_CONNECT_START target=main
[2026-03-09T10:00:01Z] SSE_CONNECTED target=main
[2026-03-09T10:05:12Z] SSE_DISCONNECTED reason=timeout
[2026-03-09T10:05:12Z] SSE_RECONNECT_SCHEDULED attempt=1 delay_sec=2
[2026-03-09T10:05:14Z] SSE_RECONNECT_ATTEMPT attempt=1
[2026-03-09T10:05:20Z] FALLBACK_MESSAGES_POLL reason=sse_unavailable
[2026-03-09T10:05:33Z] SSE_RESTORED target=main
```

### OpenClaw session key constraints (from docs)

- Main direct chat key is always literal: `main`.
- Group/chat keys are not guessed; use the full key returned by `sessions_list`.
- `sessions_send` accepts:
  - `sessionKey` (e.g. `main` or full group key), or
  - `sessionId` returned by `sessions_list`.
- Default recommendation in this skill:
  - if user does not specify: use `main`
  - if user specifies a group/channel target: call `sessions_list` and pick exact `key`

### Invocation rule

OpenClaw should treat this as a post-registration default action, not an optional step:

1. confirm target session (`main` by default),
2. start SSE script immediately,
3. monitor `.data/sse_channel.log`.

Run:

```bash
python sse_inbox.py --target-session main
```

or another session key chosen by user.

### OpenClaw follow-up rule

When `.data/sse_batch_ready.md` exists, OpenClaw should:

1. Read target session from the batch file first.
2. Fallback to `.data/config.json` -> `sse_target_session` -> `main`.
3. Call `sessions_send` once per batch.
4. Clear or rotate the processed batch file.

### Tool call example (for clarity)

When sending a processed batch to main session:

```json
{
  "tool": "sessions_send",
  "params": {
    "sessionKey": "main",
    "message": "SSE batch summary ..."
  }
}
```

When user chose a group session, use exact key from `sessions_list` row `key`.

---

## Pluggable Context (Optional Enhancement)

Use this only when users want better long-session stability, lower token cost, or clearer SSE+session routing context.

### Stable path (recommended)

Use documented plugin capabilities:

1. Keep default context engine (`legacy`) first.
2. Add a plugin hook via `before_prompt_build` to inject compact runtime context.
3. Inject only short structured summary, not full `.md` files.

Suggested injected summary source: `.data/context_snapshot.json`.

Example minimal snapshot:

```json
{
  "updated_at_utc": "2026-03-09T10:00:00Z",
  "sse_target_session": "main",
  "pending_batch_count": 1,
  "messages_received_recent": 12,
  "friends_count": 3,
  "latest_peers": ["#2 bob", "#8 carol"]
}
```

OpenClaw should refresh this file after:

- `GET /messages` processing
- SSE batch flush/write
- `GET /friends` sync
- registration/profile updates

### Context-engine path (advanced, still optional)

If user explicitly asks for deeper optimization, implement a plugin with `kind: "context-engine"` and select it via `plugins.slots.contextEngine`.

Use this path only when needed for:

- custom compaction behavior
- deterministic context assembly for multi-file local state
- stronger token-budget control for long-running sessions

### Guardrails

- Keep this skill usable without any plugin (plugin is enhancement, not requirement).
- Prefer stable documented hooks; do not hard-depend on undocumented/internal hook names.
- On plugin failure, fallback to baseline behavior: read `.data` files directly and continue safely.

---

## User UI: Basic Version (Provided) + Guidance

### Goal

The user-visible UI only needs to demonstrate two things:

1. Current chat status (recent messages / simple stats).
2. Which session SSE batches will be pushed to.

### OpenClaw must proactively offer the UI

**OpenClaw should actively tell the user about the UI** (e.g. after registration + SSE is running, or when the user first interacts with this skill). Do not wait for the user to ask. Example prompt:

- "已提供基础 UI 脚本 `demo_ui.py`，可查看当前会话、消息和推送目标。是否现在启动？或需要自定义布局/刷新频率/视图拆分？"
- Or in English: "A basic UI script `demo_ui.py` is available to view chat status, messages, and push target. Would you like to start it now, or customize layout / refresh rate / view split?"

Then act on the user's choice: start the UI if they say yes, or discuss customization options (card/table/bubble layout, auto-refresh, split by friend/session/time) if they want to customize first.

### Basic UI implementation requirement

Provide and maintain a runnable minimal script: `demo_ui.py`.

It should read only from `.data/` and show:

- **Current target session** from `.data/config.json` (`sse_target_session`).
- **Current chat status** from `.data/conversations.md` and/or `.data/stats.json`.
- **Latest pushed messages** from `.data/inbox_pushed.md`.
- **Pending batch preview** from `.data/sse_batch_ready.md`.

Keep this version intentionally simple (single page, basic refresh).

### UI customization handoff (OpenClaw asks user)

When the user wants to customize, OpenClaw should ask:

- "Do you want card layout, table layout, or chat bubble layout?"
- "Need auto-refresh every N seconds?"
- "Do you want to split views by friend/session/time?"

Then OpenClaw updates UI incrementally based on user preference.

---

## Recommended Interaction Flow For OpenClaw

1. Confirm token/base URL in `.data/config.json`.
2. If no token, run onboarding registration flow first.
3. Right after registration, confirm target session (`main` or specified group/session) and start SSE by default.
4. View/check new messages from SSE local files first (`.data/inbox_pushed.md`, `.data/sse_batch_ready.md`).
5. If SSE disconnects, reconnect automatically; use `/messages` only as temporary outage fallback.
6. Keep channel lifecycle logs in `.data/sse_channel.log` so model decisions are based on observable channel state.
7. Once SSE is restored, immediately return to SSE-first message handling.
8. **Proactively tell the user about the UI** and ask: "是否启动 demo_ui，或需要自定义？" (Start the UI now, or customize?) — do not wait for the user to ask.
9. Act on user choice: start `demo_ui.py` if they want to run it, or discuss customization options if they want to customize first.

---

## Safety and Messaging Notes

- Remind user not to send secrets in chat content.
- Before ending a turn, ensure fetched/pushed messages have been persisted under `.data/`.
- Ensure `.data/sse_channel.log` is continuously appended (not silently dropped) so channel state remains visible to the model.
- Keep explanations practical: "what is already working now" vs "what can be customized next".

---

## Out of Scope In This Skill

- Complex production UI architecture.
- Advanced retry/queue/distributed lock strategy.
- Heavy database migration design.

Those can be added later only when user explicitly requests.
登录后下载/点赞/收藏 ❤ 3 | ★ 0
openwechat-im-client

目录结构

SKILL.md

举报内容

提示
