description
Generate daily AI news digest for GGPrompts

Daily AI News Digest Generator

Generate a daily AI news digest for GGPrompts. Gather real data from multiple sources, then write to the site.

Step 0: Options Menu

Before fetching any data, use AskUserQuestion to present these options:

Question 1: "What would you like to do?"

Preview only - Fetch data and output JSON without writing files (for testing)
Generate and save - Fetch data, write files, and auto-commit+push to git
Regenerate section - Only regenerate a specific section (no auto-commit)

Question 2 (if "Regenerate section" selected): "Which section(s)?" (multi-select)

Tool Updates (Claude Code, Gemini CLI, Codex changelogs)
Trending Skills (from SkillsMP)
MCP Servers (from catalog)
Claude Code Hooks (from docs/GitHub)
TUI Tools (terminal UI tools from GitHub)
Top Stories (from feed + web search)
Trending Videos (from YouTube API)

If regenerating sections, read the existing lib/news/data/latest.json first, then only fetch/update the selected section(s).

Step 1: Fetch Data Sources

Fetch all of these in parallel:

Aggregated feed: https://personal-homepage-virid.vercel.app/api/feed
- Filter for AI/LLM/coding-related content
Web searches (run all in parallel):
- "Claude Code news" for the current month and year
- "Gemini CLI updates" for the current month and year
- "OpenAI Codex news" for the current month and year
- "AI prompt engineering news" for the current month and year
- "new MCP server site:github.com" for the current month and year
- "Claude Code hooks examples" for the current month and year
- "terminal TUI tool site:github.com" for the current month and year
Trending GitHub repos: Use trotd --json -n 20 -p gh to get today's trending repos
- Filter for AI/LLM/coding-related repos (look for keywords: ai, llm, agent, mcp, rag, prompt, coding)
- Pick 4 most relevant repos for prompt engineers
- Use the url, description, language, and stars_today fields from the output
Changelogs (fetch in parallel):
Claude Skills: https://skillsmp.com/feed.xml
- Pick 4 trending/new skills relevant to prompt engineering
- Credit SkillsMP as the source
- Set stars to 0 (SkillsMP doesn't provide repo URLs in a parseable format)
MCP Servers: Search the Docker MCP catalog using mcp-find
- Run multiple searches: "mcp" (limit 30), "ai", "database", "api", "dev"
- Look for servers with detailed descriptions indicating recent updates
- Pick 4 interesting/useful servers that would benefit prompt engineers and AI developers
- Include the install command format: mcp-add {server-name}
- These servers are part of the Docker MCP toolkit and can be dynamically enabled
YouTube Videos: Fetch from /api/youtube/trending?minDuration=180 endpoint
- First, check if the dev server is running: curl -s http://localhost:4001 > /dev/null
- If not running OR if locked (check for .next/dev/lock), kill any existing process first:
```
pkill -f "next dev" 2>/dev/null; sleep 2
npm run dev &
sleep 8  # Wait for server to be ready
```
- Use minDuration=180 (3 minutes) to filter out shorts and get substantive content
- The endpoint searches for: Claude Code, AI prompt engineering, Cursor AI, AI coding
- Returns top videos from the last 7 days sorted by view count
- Pick 4 most relevant videos for prompt engineers
- Fallback: If the API returns an error (missing key or server issues), keep the previous day's trendingVideos from latest.json, but update view counts via web search if possible
Claude Code Hooks: Search for hook examples and patterns
- Fetch official docs: https://code.claude.com/docs/en/hooks (canonical URL, avoid redirects)
- Also check: https://blog.gitbutler.com/automate-your-ai-workflows-with-claude-code-hooks
- Search GitHub for "claude code hooks" or "claude-code hooks"
- Look for hooks in categories: PreToolUse, PostToolUse, Notification, SessionStart, Stop
- Pick 2-3 interesting/useful hook examples with descriptions
- Include the hook type and a practical use case for each
TUI Tools: TUI tools rarely appear in daily GitHub trending. Use these sources instead:
- Fetch curated list: https://github.com/rothgar/awesome-tuis for comprehensive options
- Web search for recent bubbletea (Go) and ratatui (Rust) projects
- Safe defaults (well-maintained, popular): lazygit, yazi, k9s, gitui, posting, rainfrog
- Pick 3-4 tools, mixing established favorites with newer discoveries
- Include: name, description, language, stars (can estimate from awesome-tuis), GitHub URL
- Focus on tools useful for productivity, dev workflows, or beautiful terminal experiences

Step 1.5: Deduplicate Against Recent Digests

Before generating the digest, check for duplicate content from recent days:

Read recent digests: Glob for lib/news/data/2025-*.json and get the 7 most recent files (excluding today)

Extract hero URLs from all 7 days - hero stories should NEVER repeat within a week:

jq -r '.hero.sourceUrl' lib/news/data/2025-12-{08,09,10,11,12,13,14}.json 2>/dev/null | sort -u

Extract top stories/repos from last 3 days - these can repeat after 3 days if still relevant:
```
jq -r '.topStories[].sourceUrl, .trendingRepos[].url' lib/news/data/2025-12-{12,13,14}.json 2>/dev/null | sort -u
```
(Adjust the date ranges to match the most recent files)
Build exclusion sets:
- Hero exclusion set: All hero URLs from past 7 days
- Story exclusion set: Top stories and repos from past 3 days
Apply during selection:
- Hero: MUST NOT use any URL from the 7-day hero exclusion set. Pick the next best story.
- Top Stories: Avoid URLs from 3-day story exclusion set. If a story is too important to skip, reframe it with a fresh angle.
- Trending Repos: Prefer repos not featured in the last 3 days, but okay to repeat if still trending
Log duplicates found: Note any stories that were skipped due to recent coverage

This ensures each day's digest feels fresh and hero stories don't repeat within a week.

Step 2: Generate JSON

Create a JSON object matching the DailyNews interface in lib/news/types.ts.

Key sections to populate:

hero: The single most compelling story (acquisitions, major releases, security issues)
pulse: Stats about curation (stories count, sources scanned, tool updates, top 4-5 topics)
topStories: 4 most interesting stories for prompt engineers
trendingRepos: 4 relevant GitHub repos with real star counts
trendingSkills: 4 Claude Code skills from SkillsMP (include both GitHub and marketplace URLs)
newMcpServers: 3-4 new/notable MCP servers with install commands
notableHooks: 2-3 interesting Claude Code hook examples with type and description
tuiTools: 3-4 trending terminal UI tools (bubbletea, ratatui, textual ecosystem)
trendingVideos: 4 AI/coding YouTube videos (tutorials, news, podcasts)
toolUpdates: Latest versions of Claude Code, Gemini CLI, and Codex with 3 highlights each
notableQuotes: 3 interesting quotes from the sources (not marketing fluff)
allSources: All sources used, with proper attribution

Step 3: Editorial Guidelines

Hero: Pick the most compelling story - acquisitions, major releases, security issues get priority
Top Stories: Prioritize actionable news for prompt engineers and AI developers
Skills: Always credit SkillsMP (https://skillsmp.com) as the source
MCP Servers: Focus on practical tools (databases, APIs, dev tools)
Hooks: Include the hook type (PreToolUse, PostToolUse, etc.) and practical use case
TUI Tools: Highlight tools from charm.sh/bubbletea (Go) and ratatui (Rust) ecosystems. Focus on beautiful, productive terminal experiences
Videos: Prioritize tutorials, deep-dives, and genuine content over clickbait. Prefer videos from known AI/dev creators (Fireship, Matt Wolfe, AI Explained, etc.)
Quotes: Find genuine insights from developers/researchers, not PR statements
Tone: Engaging but not sensational; informative for a technical audience
Images: Use relevant Unsplash URLs for the hero image when appropriate

Step 4: Write to Site (if not preview mode)

If "Preview only" was selected:

Output the complete JSON to the conversation
Do NOT write any files
End with: "Preview complete. Run again and select 'Generate and save' to write files."

If "Generate and save" or "Regenerate section" was selected:

Read the current date and format as YYYY-MM-DD
Set generatedAt to the actual current UTC timestamp (e.g., new Date().toISOString())
Validate JSON structure before writing:
- Ensure all required fields are present (hero, pulse, topStories, trendingRepos, toolUpdates, notableQuotes, allSources)
- Verify arrays have expected lengths (4 stories, 4 repos, 3 tool updates, etc.)
- Check that all URLs are valid (no placeholders or empty strings)
Write the JSON to: lib/news/data/YYYY-MM-DD.json
Copy to: lib/news/data/latest.json
Confirm what was written

Do NOT modify lib/news/types.ts - the page reads from the JSON files.

Step 5: Verify

Run npm run build to ensure no TypeScript errors from the new data.

Step 6: Commit and Push (if "Generate and save" selected)

Only for "Generate and save" mode (not preview or regenerate section):

After the build passes, automatically commit and push the changes:

Stage the new files: git add lib/news/data/YYYY-MM-DD.json lib/news/data/latest.json

Commit with a descriptive message summarizing the digest content:

content: add daily AI news digest for {date}

- Hero: {brief hero summary}
- Top stories: {2-3 story titles}
- New MCP servers: {server names}
- Notable hooks: {hook names}
- TUI tools: {tool names}
- Skills: {skill names}

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Push to origin: git push
Confirm the push was successful

This allows the command to run unattended (e.g., before bed) and have everything committed and deployed.

Attribution Requirements

Always include these in the allSources array:

{ "title": "SkillsMP - Claude Skills Marketplace", "url": "https://skillsmp.com", "source": "skillsmp" } (if skills were included)
Any other aggregators used

Example Output Structure

{
  "date": "YYYY-MM-DD",  // Use actual current date
  "generatedAt": "YYYY-MM-DDTHH:MM:SSZ",  // Use actual current UTC timestamp
  "promptGistUrl": "https://gist.github.com/GGPrompts/5444284ae0a8f096d83c966bf22012c5",
  "modelUsed": "claude-opus-4-5-20251101",
  "hero": { ... },
  "pulse": { ... },
  "topStories": [ ... ],
  "trendingRepos": [ ... ],
  "trendingSkills": [ ... ],
  "newMcpServers": [ ... ],
  "notableHooks": [
    {
      "name": "Pre-commit Linter",
      "type": "PreToolUse",
      "description": "Runs ESLint before any file write operations",
      "sourceUrl": "https://github.com/..."
    }
  ],
  "tuiTools": [
    {
      "name": "lazygit",
      "description": "Simple terminal UI for git commands",
      "language": "Go",
      "stars": 45000,
      "url": "https://github.com/jesseduffield/lazygit"
    }
  ],
  "trendingVideos": [ ... ],
  "toolUpdates": [ ... ],
  "notableQuotes": [ ... ],
  "allSources": [ ... ],
  "prompt": "Generated via /generate-news slash command. Claude fetches data from 9 sources in parallel (aggregated RSS feed, 7 web searches, GitHub trending via trotd, SkillsMP feed, Claude Code/Gemini CLI/Codex changelogs, Docker MCP catalog via mcp-find, YouTube trending API, Claude Code hooks docs, awesome-tuis list), deduplicates against the past 7 days of digests, then curates into structured JSON."
}

GGPrompts/generate-news.md

Select an option

No results found