Labali Douyin Download User Posts

labali-douyin-download-user-posts

Treat this skill as a layered system, not a single script.

Layer Contract

1. SKILL.md (this file) is the policy layer.
   • Define goals, constraints, success criteria, and decision boundaries.
   • Stay semantic and stable across UI changes.
2. references/architecture.md is the strategy layer.
   • Define execution model, failure handling, and quality standards.
3. scripts/*.ts is the execution layer.
   • Scripts are execution assets, not the skill definition itself.
   • Use agent-browser style browser automation with Chrome CDP session reuse.
   • Keep extraction and downloading logic modular and replaceable.

Required Constraints

• Use browser automation only.
• Do not call private/undocumented Douyin APIs directly.
• Reuse manual-login session via unified Chrome CDP startup: open -na "Google Chrome" --args --remote-debugging-port=9223 --user-data-dir="$HOME/.chrome-labali-no-proxy" --no-proxy-server.
• Prefer semantic extraction from visible page state and loaded resources.
• Download target user's timeline posts and assets (text, images, optional videos).
• Export per-post post.md and media files.
• Keep partial success: keep downloaded files even if some URLs fail.
• When fixed_user_dir is provided, reuse that exact folder and do not create a new timestamp folder.
• Reuse current opened user homepage tab when it already matches target user.
• For timeline batch runs, open each post detail in a temporary tab, extract/download, then close tab and return to homepage tab.
• For each post, keep only one final video file in videos/ (prefer muxed AV or largest valid AV file).

NEVER

• Never keep multiple video files per post — select one final file (muxed AV or largest valid AV) and remove the rest.
• Never download unrelated images for a video post.
• Never report success without verifying at least one post was extracted.

Success Criteria

A run is successful only when all conditions hold:

1. A user output folder exists (new timestamp folder or provided fixed_user_dir).
2. At least one post is extracted from user timeline or direct video URL mode.
3. Per-post folders contain post.md and downloaded media when available.
4. Video posts do not download unrelated images.
5. URL output and logs retain canonical Douyin URL form.

Runtime Inputs

Use skill.yaml as the source of truth for input schema.

Operational Mode

• Default mode: guided browser flow + semantic extraction + authenticated media download.
• Links-only mode (collect_links_only=true): only collect full detail links from works timeline and export to post_links.json + post_links.txt under user output dir, no media download.
• Startup guidance:
  • launch/reuse Chrome by the unified open -na "Google Chrome" CDP command,
  • connect via CDP port,
  • open target user page,
  • check login status,
  • if required, guide user to complete manual login in the same window.
• Input guidance:
  • if user_url is missing, prompt interactively,
  • if output_dir is missing, prompt interactively with default.
• Pagination guidance:
  • scroll user timeline continuously,
  • click visible expansion controls where needed,
  • stop when no more new cards appear or explicit end markers appear.
• Text guidance:
  • prefer detail JSON desc from target postId,
  • strip platform-generated suffixes (e.g. Douyin share/promotion footers appended after the original post text).
• Publish time guidance:
  • extract from detail JSON create_time first,
  • fallback to visible page text parsing.

If failure handling or extraction decisions are unclear, load references/architecture.md. If extraction or media selection strategy is unclear, load references/plan.md.

Resources

• Architecture and standards: references/architecture.md
• Workflow map and extraction plan: references/plan.md
• Shared runtime and downloader helpers: scripts/core.ts
• Main orchestration: scripts/executor.ts
• CLI entry: scripts/run.ts
• Regression checks: tests/test_regression.sh