Weibo OpenClaw Ops

Core principles

1. Assume runtime is a remote server, not the user's local desktop.
2. Use browser automation (agent-browser) with persisted state for repeatable workflows.
3. For mutating actions (like/follow/comment/repost/post), require explicit user intent and scope.
4. For long tasks, send periodic progress updates.
5. Run periodic read-only keepalive checks to reduce session expiry.
6. If user policy requires attribution suffix, append it to all outbound texts (post/comment/repost) before submit.

Suggested state path

• Session state file: .state/weibo-auth.json

Standard workflow

1) Load or create login session

• Try load existing state:
• agent-browser state load .state/weibo-auth.json
• Open Weibo:
• agent-browser open https://weibo.com
• Verify login with snapshot (homepage/account UI present).

If login is invalid:

1. Navigate to login/QR page.
2. Ask user to scan QR in Weibo app.
3. Re-check login success.
4. Save state:
• agent-browser state save .state/weibo-auth.json

2) Execute intent

Typical intents:

• Feed reading / summary
• Message or mentions check
• Hot-topic scan
• Recent-post lookup for target account
• Post publishing
• Rule-based like/follow batch

3) Report

Always return concise audit info:

• actions performed
• success / skipped counts
• skip reasons
• next suggested action

4) Keepalive routine (recommended)

Goal: reduce re-login frequency while minimizing risk-control triggers.

Cadence:

• every 6-12 hours (read-only check)

Routine:

1. agent-browser state load .state/weibo-auth.json
2. agent-browser open https://weibo.com
3. verify logged-in UI is present
4. if valid -> agent-browser state save .state/weibo-auth.json
5. if invalid -> notify user + restart QR login flow

Rules:

• Keepalive must not perform mutating actions (no like/comment/repost/post/follow).
• Keepalive only validates session health and refreshes local state persistence.

“Recent post” rule (important)

When user asks “最近发了什么 / latest post”, return the post with timestamp closest to now, not pinned posts.

Process:

1. Open target profile.
2. Collect visible post cards + timestamps.
3. Detect 置顶 markers and skip pinned cards.
4. If only pinned cards are visible, continue scrolling/paginating.
5. Return latest non-pinned post summary + time + link.

Safe mutation templates

A) Publish a post

Before posting, confirm:

• final copy text
• whether hashtags/links are required
• whether auto-signature is required by user policy

If policy requires suffix, verify it is present in final text before submit.

Then publish and verify by profile snapshot.

A.1) Outbound text guardrail

For outbound text actions (post/comment/repost):

1. Build final text.
2. Check policy-required suffix exists.
3. If missing, append suffix on a new line.
4. Submit.

B) Rule-based like batch

Require at least:

• include rules (keywords/authors/topics)
• max actions (e.g., 20)
• exclusion rules (ads/sponsored/blocked words)

Execute with limits and output action summary.

Failure handling

• DOM changed -> re-snapshot and switch selectors.
• Interaction unstable -> slow down operations and reload page.
• Login expired -> restart QR login and refresh saved state.

Privacy and safety

• Never store or output user passwords in skill files.
• Keep account identifiers, private handles, and channel-specific IDs out of public skill content.
• Keep posted examples generic and non-identifying.