Invoke / 调用 / 呼び出し

Chinese

• "用 ai-news-digest 生成今天的 AI 资讯简报"
• "用 ai-news-digest 获取昨天的 AI 新闻"
• "用 ai-news-digest 看看前天有什么 AI 动态"
• "帮我用 ai-news-digest 整理一下最近的 AI 资讯"

English

• "Use ai-news-digest to generate today's AI news digest"
• "Use ai-news-digest to get yesterday's AI news in English"
• "Generate an AI news summary for 2026-01-20 using ai-news-digest"
• "Use ai-news-digest to summarize recent AI developments"

Japanese

• "ai-news-digest で今日のAIニュース要約を日本語で作成して"
• "ai-news-digest で昨日のAIニュースを取得して"
• "ai-news-digest を使って最近のAI動向をまとめて"
• "ai-news-digest で一昨日のAI関連ニュースを教えて"

Objectives

• Deliverables: 1 Markdown digest (JSON data optional)
• Each news item: Title, time, source, link, 1-3 sentence summary, tags (optional "why it's important")
• Default strategy: RSS/Atom first, HTML fallback; do not bypass paywalls
• Default output: Chinese (if source is in English, translate to Chinese before output)

Inputs (Clarify First)

• Time window (default "today"):
  • Natural language:

    今天/昨天/前天

    (extendable to

    过去7天/最近24 hours

    )
  • Specified date:

    YYYY-MM-DD

    (interpreted as "00:00-23:59:59 on that day" in the user's time zone)
  • Specific start/end time:

    since/until

    (with time zone)
• Time zone: Default

  UTC+8

  (recommended to implement default

  Asia/Shanghai

  ), users can explicitly specify
• Output language: Default

  zh

  (Chinese); optional

  en

  /

  both

• Topic scope: Research/Product/Open Source/Investment & Financing/Policy (multiple selection allowed)
• Source scope: Use default list or user-specified subset (see

  references/sources.md

  )
• Output format: Markdown (default) / JSON / both
• Maximum number of items: Total limit and per-topic limit (default total 20)
• Processing depth: Title only with summary (default) / attempt full text extraction (for non-paywalled content)

Workflow (Recommended Architecture)

1. Load source list (RSS first; adapt HTML if RSS is unavailable)
2. Fetch and standardize items (title/link/publish time/summary/source)
3. Deduplication and merging (same link/similar title/cross-source duplicates)
4. Topic classification and sorting (time + source weight + keywords)
5. Generate digest (apply template

   assets/digest-template.md

   , attach source links)
6. Quality check (time window hit, no duplicates, each item has link/source)

Guardrails (Compliance & Stability)

• Do not bypass paywalls/login restrictions; only use public title/summary/meta info for paywalled content
• Respect site terms and robots.txt; downgrade to "only fetch RSS items" if necessary
• Degradable on failure: Only output the "retrievable" part and list failed sources with reasons
• Translation/summary must retain traceability: Always keep original link and source in output; retain original

  title_raw/summary_raw

  in JSON if necessary
• Confirm output path and overwrite permission before any file writing/overwriting export

How to Run

CLI Commands

# Enter script directory
cd skills/public/ai-news-digest/scripts

# Get today's news digest
python run.py --day 今天

# Get yesterday's news
python run.py --day yesterday

# Disable SSL certificate verification if encountering local SSL certificate chain issues (not recommended)
python run.py --day yesterday --insecure

# Specify date
python run.py --day 2026-01-15

# Output in JSON format
python run.py --day 今天 --format json

# Export as image (Pillow required)
python run.py --day 今天 --format image

# Export as landscape image (suitable for official accounts)
python run.py --day 今天 --format image --image-preset landscape

# Export as light theme image
python run.py --day 今天 --format image --image-theme light

# Write to file
python run.py --day 今天 --out digest.md

# Use LLM for translation (API key required)
python run.py --day 今天 --llm

# Detailed output
python run.py --day 今天 --verbose

# Run smoke test
python run.py --test

CLI Parameters

--day, -d

--since

--until

--tz

--lang, -l

--format, -f

--image-preset

--image-theme

--out, -o

--topics, -t

--sources, -s

--max

--max-per-topic

--llm

--verbose, -v

--insecure

Script Module Description

run.py

time_window.py

fetch_feeds.py

parse_feeds.py

dedupe.py

classify_rank.py

render_digest.py

render_image.py

summarize_llm.py

Dependencies

• Python 3.10+
• No third-party dependencies required for Markdown/JSON output

• Pillow

  : Image rendering functionality (

  --format image

  )

• pyyaml

  : More complete YAML parsing (the script has a built-in simplified parser, so

  sources.yaml

  can be loaded normally without installation)

• anthropic

  or

  openai

  : LLM translation functionality

Install Optional Dependencies

# Install image rendering support
pip install Pillow

# Install all optional dependencies
pip install Pillow pyyaml anthropic

Note:

• If Pillow is not installed, Markdown and JSON output work normally, only image output is unavailable
• If pyyaml is not installed, the script uses the built-in simplified YAML parser, and the complete source list can be loaded normally

Resources

• Data model and output specification:

  references/output-spec.md

• Source registry:

  references/sources.yaml

• Source evaluation description:

  references/sources.md

• Time window specification:

  references/time-window.md

• Topic keywords:

  references/topic-keywords.md

• Translation specification:

  references/translation.md

• Markdown template:

  assets/digest-template.md

• LLM prompt:

  assets/summarize-prompt.md