WeChat Video Channel Publish

This skill packages a focused 微信视频号 publishing workflow for local agent execution.

Use it when the task is specifically about:

• 视频号扫码登录
• 上传短视频
• 设置标题、标签、正文长描述
• 设置封面图
• 定时发布
• 保存草稿（best effort；平台 UI 可能在部分状态下拒绝）
• 发布后补评论

Quick Start

1. Read references/runtime-requirements.md.
2. Build the CLI if dist/cli.js does not exist.
3. Run tencent login or tencent check.
4. Run tencent upload with absolute video and cover paths.
5. If the task uses a publishing queue, first normalize it into direct CLI arguments.
6. Before any public repo push, read references/privacy-checklist.md.

Workflow

Step 1: Prepare Runtime

• Run all commands in the repo root.
• If needed, execute:
  • npm install
  • npx playwright install chromium
  • npm run build

Step 2: Login or Reuse Cookie

• For first-time login:
  • node dist/cli.js tencent login --account <name>
• For validation:
  • node dist/cli.js tencent check --account <name>

The cookie file is stored outside the repo by default and must never be committed.

Step 3: Publish a Video

• Read references/publish-workflow.md.
• Use absolute paths for --file and --cover.
• Supported fields:
  • --title
  • --tags
  • --caption
  • --cover
  • --pinned-comment
  • --schedule
  • --category
  • --draft

Step 4: Validate Carefully

• Prefer --draft when verifying a new selector path or runtime change, but do not combine it with --schedule.
• Treat cover and pinned-comment as runtime-sensitive features.
• If UI changes break these steps, keep login and upload logic intact and narrow the failure to the affected selector flow.

Step 5: Repo Hygiene

• Keep cookies, screenshots, local logs, and temporary exports out of git.
• When sharing publicly, include only reusable code, templates, and references.

Resource Loading Rules

• Read references/runtime-requirements.md before execution.
• Read references/publish-workflow.md when mapping a publishing plan into CLI arguments.
• Read references/privacy-checklist.md before creating or updating a public repository.

Output Expectations

• The agent should execute the real CLI when feasible.
• The agent should report what fields were actually attempted at runtime.
• If a selector-dependent step is flaky, the agent should say exactly which sub-step failed:
  • login
  • upload
  • caption
  • cover
  • comment

Guardrails

• Do not print or commit cookie JSON.
• Do not embed user-specific absolute paths in committed docs except as obvious placeholders.
• Do not claim comment pinning is fully reliable unless it was verified in the current session.
• Do not claim --draft + --schedule is supported; use --schedule without --draft for real scheduled publishing.