article-summary-card（文章摘要卡）

Article Summary Card

Turn an article into a concise summary bundle.

When to use

• Summarize a webpage or article and deliver the result as an image.
• Convert long text into a reusable set of outputs: JSON, Markdown, HTML, PNG.
• Produce repeatable summary cards with a consistent layout and predictable sizing.

Runtime Requirements

• python3
• curl
• Chrome or Chromium for headless screenshots
• Python package: Pillow

If Chrome is not installed at the default path, adjust the browser candidate list in scripts/render_card.py.

Workflow

1. Read the input article from a URL or local file.
2. Extract the title and article body; remove obvious page chrome when possible.
3. In the current session, run two prompt rounds:

- Round 1: create a summary plan that decides how the article should be divided into sections. - Round 1 must also generate 3 to 8 short tags for the article. - Round 2: write the final summary JSON according to that plan.

1. Use the unified renderer to export the final summary JSON as Markdown, HTML, and PNG.
2. Verify the outputs exist and have a reasonable size.

Commands

Extract article text for the session workflow:

python3 article-summary-card/scripts/extract_article.py \
  --url 'https://example.com/article' \
  --out output/article-input.json

The extracted JSON contains:

• title
• source
• article_text

Then in the current session:

• Use references/prompts/plan-system.md and references/prompts/plan-user.md to design the summary structure.
• Use references/prompts/summary-system.md and references/prompts/summary-user.md to write the final summary JSON.
• Include tags in the final summary JSON and show them at the end of the rendered card and Markdown output.

Preferred final export:

python3 article-summary-card/scripts/render_outputs.py \
  --summary output/article-summary.json \
  --out-stem output/article-summary

This produces:

• output/article-summary.md
• output/article-summary.html
• output/article-summary.png

Optional lower-level renderers:

python3 article-summary-card/scripts/render_markdown.py --summary output/article-summary.json --out output/article-summary.md
python3 article-summary-card/scripts/render_card.py --summary output/article-summary.json --out output/article-summary.png

Adjust styles in:

article-summary-card/assets/templates/mobile-card.css

The renderer keeps HTML and CSS separate:

article-summary-card/assets/templates/mobile-card.html
article-summary-card/assets/templates/mobile-card.css

The size system is based on a 375px design width multiplied by SCREEN_RATIO in CSS and Python.

Optional helper: generate a local heuristic draft JSON when you want a quick bootstrap, but do not treat it as the preferred path for high-quality output:

python3 article-summary-card/scripts/summarize_article.py \
  --url 'https://example.com/article' \
  --out output/article-summary-draft.json \
  --mode heuristic

For final output, replace or rewrite that draft in-session and then use render_outputs.py. summarize_article.py is a compatibility helper, not the main summarizer.

Cross-Platform Adapters

• Codex

- Native entrypoint is this skill folder itself: article-summary-card/SKILL.md - Optional UI metadata: article-summary-card/agents/openai.yaml

• Claude Code

- Project slash command: .claude/commands/article-summary-card.md - Usage pattern: /article-summary-card <url-or-file> [output-stem]

• OpenClaw

- OpenClaw uses skill folders containing SKILL.md, so this same directory is portable. - Install helper:

python3 article-summary-card/scripts/install_openclaw.py

- Default destination: ~/.openclaw/workspace/skills/article-summary-card

Notes

• This skill prefers deterministic rendering over image-generation models so long Chinese text stays accurate.
• The preferred summarizer is the current session model, not an API call inside Python.
• Summary instructions are intentionally extracted into references/prompts/ so they can be revised without editing Python code.
• Cross-platform portability comes from keeping one shared skill core and only adding thin platform entrypoints.
• If a site is hard to extract, inspect the HTML and add a site-specific extraction rule in scripts/summarize_article.py.
• For very long articles, keep the summary short enough to fit on one card. If it still overflows, shorten section points before re-rendering.
• summarize_article.py no longer performs LLM calls; it only generates a heuristic draft JSON.
• When DOM height measurement succeeds, the renderer trusts that height and skips whitespace auto-cropping to avoid cutting off low-contrast tags or footer content.
• The renderer uses overscan-then-crop for long screenshots to avoid incomplete bottom rendering in headless Chrome.