Claude Skill /unzuck 소셜 미디어 피드 수집기

Take back your feeds. Scans 7 platforms, kills the noise, keeps the signal. unzuck is a Claude Code plugin that scans your social media feeds, scores content against your interests, deep-dives on the best items, and generates a beautiful interactive HTML dashboard — your daily content digest, on autopilot. - Scans 7 platforms in parallel — YouTube, Twitter/X, Reddit, Hacker News, LinkedIn, Facebook, Instagram - Scores every item against your personal interest profile (weighted topics, negative topics, engagement signals) - Deep-dives on the top items — fetches full articles, transcripts, threads, and synthesizes detailed summaries that teach you the core insight without clicking through - Generates an interactive dashboard — filterable by platform, tags, and score, with dark/light theme, search, and embedded media - Evolves your profile over time — detects emerging interests, flags stale topics, suggests new subreddits /plugin marketplace add donttalkaboutit/unzuck /plugin install unzuck@unzuck claude --chrome /unzuck Or say any of these to Claude: - "unzuck my feeds" - "curate my feeds" - "what's new on my feeds" - "daily digest" - "scan my feeds" On your first run, unzuck walks you through a ~2 minute onboarding: - Choose where to save digests (e.g., ~/Documents/content-digests/ ) - Pick which platforms to scan and how many items per platform - Tell it your interests — or let it bootstrap by scanning your feeds first - Review your profile and suppress topics you never want to see After setup, every subsequent run is fully autonomous — no confirmations, no interruptions, just a finished dashboard. Don't want to list interests manually? Say "bootstrap from my feeds" — unzuck will scan your YouTube topic chips, Twitter timeline, Reddit subscriptions, and more to infer your interest profile automatically. - Claude Code with the Claude in Chrome extension - Logged into the platforms you want scanned (in your Chrome browser) - Hacker News and Reddit work without login /unzuck (SKILL.md) │ ├── Step 0: Load/bootstrap interest profile ├── Step 1: Pre-flight — check which platforms are logged in ├── Step 2: Parallel scan — one platform-scanner agent per platform ├── Step 2.5: Dedup (cross-platform + cross-session via history.json) ├── Step 3: Score all items (0-100) against interest profile ├── Step 4: Parallel deep dives — one deep-dive-agent per top item ├── Step 5: Generate HTML dashboard from template ├── Step 5.5: Profile refinement — suggest emerging/stale topics └── Step 6: Save history, present results | Agent | Purpose | |---|---| platform-scanner | Navigates a single platform via Chrome, scrolls, extracts items with rich descriptions and engagement data | deep-dive-agent | Fetches full content (articles, transcripts, threads), enriches from supplementary sources, produces teach-me-level summaries | Items are scored 0-100 based on: - Topic relevance — weighted match against your interest profile - Recency — +5 for content from the last 24 hours - Engagement — +3 for viral content (300+ HN points, 100K+ YouTube views, etc.) - Content quality — +5 for rich, detailed descriptions - Negative topics — configurable penalty (default -30) for topics you want suppressed - History — -20 for items you've already seen in previous runs The generated HTML dashboard includes: - Platform filters and tag cloud - Score slider to set minimum threshold - Search across titles and descriptions - Dark/light theme toggle - Embedded YouTube players and Instagram embeds - Expandable deep-dive sections with key takeaways and timestamped insights - Keyboard navigation (arrow keys, enter to expand, b to bookmark) Your profile lives in interests.json and includes: { "topics": [ { "name": "AI/ML", "weight": 0.9, "keywords": ["LLM", "transformer"] } ], "negativeTopics": [ { "name": "Crypto trading", "keywords": ["altcoin"], "penalty": -30 } ], "platformOverrides": { "hackernews": { "maxItems": 25 }, "youtube": { "maxItems": 10 }, "facebook": { "maxItems": 0 } } } - Weights: 0.8-1.0 = core interest, 0.5-0.7 = solid, 0.2-0.4 = casual - Negative topics: push matching items down or off the dashboard - Platform overrides: control item count per platform (0 = skip) The profile evolves automatically — unzuck tracks topic match rates across runs and suggests additions, weight adjustments, and new subreddits. | Platform | Login required | Default items | |---|---|---| | Hacker News | No | 25 | | No (public) / Yes (personalized) | 15 | | | YouTube | Yes (Google) | 10 | | Twitter/X | Yes | 10 | | Yes | 5 | | | Yes + opt-in | Off | | | Yes + opt-in | Off | | File | Purpose | |---|---| interests.json | Your interest profile (persistent) | history.json | Seen URLs for cross-session dedup (persistent) | items-scored.json | Scored items for current run (overwritten) | content-digest-YYYY-MM-DD.html | Daily dashboard output | MIT