Jason CS Lin

Synanews

演算本日關鍵產業新聞的拓樸。

Repository Target

This project is configured as Synanews and pushes generated artifacts to the GitHub Pages repository for user lcslin.

• GitHub repository: lcslin/lcslin.github.io
• Current git remote: origin -> https://github.com/lcslin/lcslin.github.io.git
• Branch: main

What This Project Does

gen-daily-graph.py fetches headlines from https://www.itis.org.tw/member/DailyNews.aspx, sends a numbered title list to an LLM, writes the returned graph to a dated JSON file, and embeds that day’s data into daily-graph.html.

The current implementation is headline-driven. It does not scrape article summaries or full article content.

Inputs

Environment variables are loaded from .env with python-dotenv.

Required:

• TAITRA_API_KEY

Optional:

• TAITRA_BASE_URL default: https://runway.taitra.ai
• MODEL default: claude-sonnet
• GIT_PUSH_BRANCH for commit.sh
• GIT_COMMIT_MESSAGE for commit.sh
• GIT_DEPLOY_KEY_PATH for commit.sh

Recommended secret file location for cron:

• ~/.config/synanews.env

Outputs

Running the generator produces:

• YYYY-MM-DD.json: graph data for the selected day
• daily-graph.html: HTML generated from daily-graph-template.html with that day’s data embedded

The JSON normally includes:

• nodes
• links
• newsDatabase
• titleToUrl

CLI

python gen-daily-graph.py
python gen-daily-graph.py --today
python gen-daily-graph.py --date 20260206

Behavior:

• Default or --today: fetch DailyNews.aspx
• --date YYYYMMDD: fetch DailyNews.aspx?Day=YYYYMMDD
• Output JSON filename follows the selected date, for example 2026-02-06.json

-o/--output is not supported by the current implementation.

Implementation Notes

Scraping

• Uses requests.Session()
• Parses elements with class dailynews-link-item
• Stores each item as title, url, and content
• In the current code, content is the same as title

LLM Input

• The LLM receives a numbered list of titles only
• The system prompt is loaded from entity-graph-prompt.md
• The model is expected to return valid JSON

Post-processing

• URLs are backfilled into newsDatabase
• A titleToUrl mapping is added to the output JSON
• HTML is rendered by replacing placeholders in daily-graph-template.html

Supporting Scripts

cron-entry.sh

• Loads ~/.config/synanews.env if present
• Loads .env
• Runs gen-daily-graph.py
• Runs commit.sh

commit.sh

• Stages *.html and *.json
• Creates a git commit if there are changes
• Pushes to lcslin/lcslin.github.io
• Loads ~/.config/synanews.env before repo-local .env
• Auth order is: deploy key, GITHUB_TOKEN or GH_TOKEN, then gh auth token, then plain git remote auth

Cronjob Auth

For local cronjob execution, the preferred setup is to keep secrets outside the repo and let the scripts load them from a fixed path.

Recommended ~/.config/synanews.env content:

TAITRA_API_KEY=...
TAITRA_BASE_URL=https://runway.taitra.ai
GITHUB_TOKEN=...

Recommended crontab entry:

0 8 * * * /Users/cslin/app-dev/Synanews/cron-entry.sh >> /Users/cslin/app-dev/Synanews/cron.log 2>&1

commit.sh still supports gh auth token as a fallback, but a fixed env file is more predictable for cron.

Known Gaps

• Login redirects only emit a warning in the current Python script; they do not fail fast
• The project depends on daily-graph-template.html, which must exist beside the Python script
• The prompt mentions source inference, but runtime input is still headline-only
• The secret env file must stay outside the repo and should use restrictive permissions such as chmod 600