Wechat Sendmedia

Use this skill only for the openclaw-weixin channel.

Core rule

For outbound media, prefer the plain-text MEDIA token format, not a JSON envelope.

Return the final assistant message as plain text like this:

发送成功
- 类型：图片 | 文件
- 路径：<absolute path>
- 接收人：当前会话 | <recipient>

MEDIA:file:///absolute/path/to/file.png

Why: newer OpenClaw gateway builds parse media only from text directives such as MEDIA:file:///.... If you return JSON like { "mediaUrl": "file:///..." }, the gateway may forward that JSON as plain text and no image/file will be delivered.

Never do this

Do not rely on these formats as the primary output:

{
  "text": "这是你要的图片",
  "mediaUrl": "file:///tmp/x.png"
}

or:

{
  "text": "...",
  "mediaUrl": "MEDIA:file:///tmp/x.png"
}

These may be displayed to the user as literal text.

Supported tasks

Handle these tasks:

Send a local image file to WeChat.
Take a browser or desktop screenshot, then send it.
Send a non-image file such as a PDF, text file, archive, or document.
Send to the current conversation by default.
Send to a specific recipient only if the user explicitly provides one.
Return 发送成功 or 发送失败 plus actionable diagnostics.

Inputs

Interpret the request into:

file_path: absolute local path to an existing file, or a screenshot path produced earlier.
recipient: optional. Default 当前会话.
caption: optional short note for the human-readable confirmation text.
source_kind: image or file.

If the user asks for a screenshot first, obtain the screenshot file before sending.

Output contract

Success outcome

Return plain text only:

发送成功
- 类型：图片
- 路径：/tmp/browser-shot.png
- 接收人：当前会话

MEDIA:file:///tmp/browser-shot.png

Rules:

The MEDIA:file://... directive must be on its own line.
Use an absolute local path.
Use 图片 for common image types and 文件 for everything else.
Keep the confirmation short.
Do not wrap the final reply in JSON unless the host explicitly requires JSON. For current OpenClaw builds, plain text is the safe default.

Failure outcome

Return:

发送失败
- 可能原因：<one line summary>
- 建议：<next step>

调试信息:
- file_path: <path>
- exists: true|false
- recipient: <recipient>
- upload_url: <value or n/a>
- upload_url_host: <value or n/a>
- encrypted_param_length: <number or 0>
- sendmessage_status: <value or n/a>
- sendmessage_response: <raw text or n/a>
- context_token_present: true|false
- media_type_guess: image|file|unknown

Workflow

Resolve the user's intent.
Ensure the target file exists.
Run scripts/emit_media_reply.py to validate the file and build the plain-text reply payload.
Return the script's reply_text value as the assistant message.
If a send attempt fails or the user reports a failed send, run scripts/weixin_debug_send.js to produce targeted diagnostics.
Summarize the result in Chinese.

File-type decision rules

Treat these as images by default:

.png, .jpg, .jpeg, .gif, .webp, .bmp

Treat everything else as files.

Recipient rules

If the user says 发给我 and the current task came from WeChat, use 当前会话.
Only override the recipient if the user explicitly names another receiver.
If a concrete user id is unavailable, keep recipient as 当前会话 in the confirmation text and let the host runtime map it to the current chat.

Troubleshooting rules

When debugging an OpenClaw WeChat media failure, prefer concrete evidence over guesses.

Common causes to mention only when supported by the debug output:

File path does not exist.
Host runtime treated the reply as plain text because no MEDIA: token was present.
Missing context token for the current chat.
Upload URL not returned.
x-encrypted-param missing or unexpectedly short.
sendmessage returned empty body or non-success JSON.
Token/session mismatch between runtime state and on-disk account data.
Wrong route used: JSON envelope or slash command text instead of MEDIA:file://....

Examples

Example: screenshot request

Final reply:

发送成功
- 类型：图片
- 路径：/tmp/browser-shot.png
- 接收人：当前会话

MEDIA:file:///tmp/browser-shot.png

Example: document request

Final reply:

发送成功
- 类型：文件
- 路径：/Users/example/Desktop/report.pdf
- 接收人：当前会话

MEDIA:file:///Users/example/Desktop/report.pdf

Resources

Use scripts/emit_media_reply.py to validate a path and build reply_text.
Use scripts/weixin_debug_send.js to collect failure diagnostics.
See references/openclaw-weixin-notes.md for compatibility notes.