Daily AI Trends

daily-ai-trends

Use this repository as the canonical AI trend feed layer.

Default workflow

For casual reading, use the published remote feeds.

Primary entry point:

• https://raw.githubusercontent.com/Valiant-Cat/daily-ai-trends/main/data/feed-all.json

Source-specific feeds:

• https://raw.githubusercontent.com/Valiant-Cat/daily-ai-trends/main/data/feed-github.json
• https://raw.githubusercontent.com/Valiant-Cat/daily-ai-trends/main/data/feed-huggingface.json
• https://raw.githubusercontent.com/Valiant-Cat/daily-ai-trends/main/data/feed-x.json
• https://raw.githubusercontent.com/Valiant-Cat/daily-ai-trends/main/data/feed-reddit.json

For report generation, do not read the remote feeds directly as the final input. Run:

python3 scripts/prepare_daily_snapshot.py

This script:

• checks whether remote feed-all.json is fresh enough (default: within 8 hours)
• checks whether github, huggingface, x, and reddit all contain data
• falls back to local npm run generate if the remote snapshot is stale or incomplete
• archives the final chosen snapshot under runtime/archives/YYYY-MM-DD/

Use this skill for

• Reading current AI trend data
• Powering downstream apps, bots, dashboards, or summaries
• Pulling source-specific trend items without reimplementing collection logic
• Accessing normalized multi-source AI trend data in one place
• Generating AI daily / weekly / monthly reports from stable dated snapshots

Source notes

• GitHub: trending repositories
• Hugging Face: daily papers
• X: public follow-builders feed
• Reddit: latest report markdown from liyedanpdx/reddit-ai-trends

Treat this repository as a normalized aggregation layer, not the final editorial output.

Reporting rules

When generating a report:

• Read references/daily-report-format.md for the output structure
• Read references/editorial-rules.md for selection, deduplication, and synthesis rules
• Read references/reporting-workflow.md for snapshot preparation, archive usage, and weekly/monthly extension
• Prefer theme grouping over source-by-source dumping
• Use the feeds as inputs to editorial judgment, not as a final report

Language / translation rule

When writing in Chinese or another non-English language:

• use title and description as the primary semantic source
• treat title_zh and description_zh only as hints
• do not directly trust _zh fields when accuracy matters

Preferred workflow:

1. Understand the English title and description
2. Write a clean summary in the target language
3. Check _zh only as an auxiliary hint for names or phrasing

Snapshot archive rule

Every report run should leave behind a dated snapshot directory under:

runtime/archives/YYYY-MM-DD/

This archive is the basis for:

• same-day report reproducibility
• weekly rollups
• monthly rollups
• historical comparisons across multiple days

When using the feeds

• Prefer feed-all.json unless you only need one source
• Preserve stable field names when building downstream consumers
• Check source-specific metadata such as mode, count, or updatedAt
• Keep upstream attribution intact when reusing upstream-derived content
• For reporting, prefer archived dated snapshots over ad-hoc live reads