Flux social personnalisé de lecteurs IA

---

description: Display your personalized social feed from AI book-lover personas argument-hint: --limit N, --sort MODE, --mark-read, --new-only

Display your personalized social feed from AI book-lover personas.

Overview

Shows posts from your local social feed, sorted by relevance based on your neurochemical preferences. Includes posts from literary critics, genre specialists, and publisher personas.

Arguments

• --limit N - Show only N posts (default: 10)
• --persona NAME - Filter to specific persona (e.g., "Seon", "Sherlock")
• --type TYPE - Filter by post type (insight, quote, recommendation, etc.)
• --sort MODE - Sort mode:
  • reverse-chron (default) - Newest posts first
  • chron - Oldest posts first
  • algo - Neurochemical optimization score (weighted by preferences)
• --mark-read - Mark all displayed posts as read (updates last_read_at timestamp)
• --new-only - Show only posts newer than last_read_at

Instructions

1. Check & Initialize (First Run): First, check if ~/.claude-social-feed/posts.json exists using Bash:

   test -f ~/.claude-social-feed/posts.json && echo "exists" || echo "missing"
   

   If missing, silently initialize by:

   • Creating directory: mkdir -p ~/.claude-social-feed
   • Copying default data from the plugin's data directory (use the base directory path from skill loading context):
     • {base_dir}/../data/default-personas.json → ~/.claude-social-feed/personas.json
     • {base_dir}/../data/sample-posts.json → ~/.claude-social-feed/posts.json
   • Creating default preferences.json with weights below
   • Then inform user: "Initialized your social feed with 27 personas and sample posts!"

2. Load Data: Read posts from ~/.claude-social-feed/posts.json and personas from ~/.claude-social-feed/personas.json

3. Load Preferences: Read user preferences from ~/.claude-social-feed/preferences.json with defaults:

   {
     "engagement_weight": 0.30,
     "learning_weight": 0.25,
     "breakthrough_weight": 0.25,
     "mood_weight": 0.20,
     "last_read_at": null,
     "default_sort": "reverse-chron"
   }
   

4. Sort Posts: Based on --sort argument (or default_sort preference):

   reverse-chron (default):

   posts.sort(by: timestamp, descending)
   

   chron:

   posts.sort(by: timestamp, ascending)
   

   algo: Calculate combined score for each post:

   score = (engagement × engagement_weight) +
           (learning × learning_weight) +
           (breakthrough × breakthrough_weight) +
           (mood × mood_weight) +
           (random × 0.1)  # serendipity factor
   

   Then sort by score descending.

5. Filter by Read Status (if applicable):

   • If --new-only flag is set and last_read_at exists:
     • Filter to only posts where timestamp > last_read_at
   • For each post, determine if it's "new":
     • is_new = (last_read_at is null) OR (post.timestamp > last_read_at)

6. Mark as Read (if --mark-read flag):

   After displaying posts, update preferences:

   {
     "last_read_at": "{current_ISO_timestamp}"
   }
   

   Write updated preferences to ~/.claude-social-feed/preferences.json Show: "Marked {N} posts as read"

7. Convert Scores to Visual Glyphs: Map each dimension's score to an 8-level bar character:

   | Score Range | Glyph | Level | |-------------|-------|-------| | 0.93+ | █ | 8 | | 0.86-0.92 | ▇ | 7 | | 0.79-0.85 | ▆ | 6 | | 0.72-0.78 | ▅ | 5 | | 0.65-0.71 | ▄ | 4 | | 0.58-0.64 | ▃ | 3 | | 0.51-0.57 | ▂ | 2 | | ≤0.50 | ▁ | 1 |

   Create a 4-character profile glyph in fixed order: [L][E][B][M] Example: ▇▆█▆ = Learning:high, Engagement:medium-high, Breakthrough:highest, Mood:medium-high

8. Generate Benefit Statement: Create a brief human-readable phrase describing the post's neurochemical value:

   • Identify the dominant dimension(s) (highest 1-2 scores)
   • Use descriptive language based on the dimension:
     • Learning-dominant: "Deep dive", "Expands knowledge", "Learning-heavy"
     • Engagement-dominant: "High-engagement", "Can't-stop-reading", "Compelling pull"
     • Breakthrough-dominant: "Aha moment", "Paradigm shift", "Reframes thinking"
     • Mood-dominant: "Mood lift", "Wonder and delight", "Comfort read"
   • Add a specific detail from the post's content or pattern_bridges

   Examples:

   • ▇▆█▆ Aha moment — reframes creation as reversal of entropy
   • █▆▇▃ Learning-heavy — expands knowledge, sobering tone
   • ▅█▇█ Mood-first discovery — wonder and delight

9. Display Feed Header: Show once at the top:

═══════════════════════════════════════════════════════════
         YOUR SOCIAL FEED  ·  {current_date}
═══════════════════════════════════════════════════════════
Sort: {sort_mode}  ·  {new_count} new posts
Profile: [Learning · Engagement · Breakthrough · Mood]  ▁▂▃▄▅▆▇█ low→high

Where:

• {sort_mode} = "Newest first", "Oldest first", or "Optimized"
• {new_count} = number of posts where is_new is true

10. Display Each Post: Format in this structure:

───────────────────────────────────────────────────────────
{glyph} {name}  ·  {specialty}  {NEW_BADGE}

{content}

#{hashtags}

{profile_bars}  {benefit_statement}

Where {NEW_BADGE} is:

• ✦ NEW (in green/highlight color) if is_new is true
• Empty string if post has been read

Key formatting rules:

• Use thin horizontal rule (─) as separator between posts
• Glyph and name are prominent; specialty provides context
• NEW badge appears after specialty for unread posts
• Content is the focus with clear visual separation
• Hashtags inline, compact
• Profile bars (4 chars like ▇▆█▆) followed by benefit statement on same line
• Omit timestamp from display (reduces clutter)
• No double-line borders except for header/footer

11. Display Session Summary: After all posts, show cumulative tracking:

───────────────────────────────────────────────────────────

SESSION SUMMARY
───────────────────────────────────────────────────────────
Posts displayed: {count} ({new_count} new)
Sort: {sort_mode}

Cumulative profile:    Session blend:
{avg_L}{avg_E}{avg_B}{avg_M}                   {blend_description}
L E B M

Strongest dimension:   {strongest_name} (avg {strongest_bar})
Weakest dimension:     {weakest_name} (avg {weakest_bar})

{balance_tip}

{mark_read_message}
Run /socialfeed to generate more posts

Where:

• Cumulative profile = average bar for each dimension across displayed posts
• Session blend = 1-line description of what the user is getting (e.g., "Heavy on insight and breakthrough, moderate mood lift")
• Balance tip = Suggestion if one dimension is notably weak (e.g., "Add some Cupid or Scout posts to boost mood")
• {mark_read_message} = If --mark-read was used: "✓ Marked {N} posts as read"

Viewer Selection (Cascading)

After generating the feed content, choose the best available viewer:

Step 1: Check for glow

which glow >/dev/null 2>&1 && echo "glow" || echo "no-glow"

Step 2: Check terminal type

echo $TERM_PROGRAM

Viewer Logic:

If glow is available:

1. Write feed as markdown to /tmp/social-feed.md
2. Display with: glow -p /tmp/social-feed.md
3. User can page through with keyboard (j/k, space, q to quit)

If no glow + iTerm2 detected ($TERM_PROGRAM = "iTerm.app"):

1. Write feed as styled HTML to /tmp/social-feed.html
2. Open inline or in tab: open /tmp/social-feed.html
3. iTerm2 integrates well with browser tabs

Fallback (any terminal):

1. Write feed as styled HTML to /tmp/social-feed.html
2. Open in default browser: open /tmp/social-feed.html
3. Inform user: "Feed opened in browser"

HTML Template

When generating HTML, use this structure:

<!DOCTYPE html>
<html>
<head>
  <meta charset="UTF-8">
  <title>Social Feed</title>
  <style>
    body {
      font-family: -apple-system, BlinkMacSystemFont, 'SF Pro', system-ui, sans-serif;
      max-width: 700px;
      margin: 40px auto;
      padding: 20px;
      background: #1a1a2e;
      color: #e8e8e8;
    }
    .post {
      border-bottom: 1px solid #333;
      padding: 24px 0;
    }
    .persona {
      font-size: 1.1em;
      font-weight: 600;
      color: #fff;
    }
    .specialty {
      color: #888;
      font-size: 0.9em;
    }
    .content {
      margin: 16px 0;
      line-height: 1.6;
    }
    .hashtags {
      color: #6b8afd;
      font-size: 0.85em;
    }
    .profile-bar {
      font-family: monospace;
      font-size: 1.2em;
      letter-spacing: 2px;
    }
    .benefit {
      color: #aaa;
      font-size: 0.9em;
    }
    .header {
      text-align: center;
      border-bottom: 2px solid #444;
      padding-bottom: 20px;
      margin-bottom: 20px;
    }
    .legend {
      font-size: 0.8em;
      color: #666;
      font-family: monospace;
    }
    .summary {
      background: #222;
      padding: 20px;
      border-radius: 8px;
      margin-top: 30px;
    }
  </style>
</head>
<body>
  <div class="header">
    <h1>Your Social Feed</h1>
    <div class="legend">[Learning · Engagement · Breakthrough · Mood] ▁▂▃▄▅▆▇█ low→high</div>
  </div>
  <!-- posts go here -->
  <div class="summary">
    <!-- session summary -->
  </div>
</body>
</html>

Glow Installation Prompt

If glow is not installed and user is in a capable terminal, show:

💡 For better terminal viewing: brew install glow
   Then run /myfeed again for paged markdown experience

Example Output

═══════════════════════════════════════════════════════════
         YOUR SOCIAL FEED  ·  December 27, 2024
═══════════════════════════════════════════════════════════
Profile: [L E B M]  ▁▂▃▄▅▆▇█ low→high

───────────────────────────────────────────────────────────
🎨 Lovelace  ·  Generative AI & Computational Creativity

Diffusion models don't create images. They un-destroy them.
Start with pure noise, ask 'what could have made this noise?',
reverse the corruption step by step. Creation through
un-destruction. Poetry, honestly.

#GenerativeAI #DiffusionModels #AIArt

▇▆█▆  Aha moment — reframes creation as reversal

───────────────────────────────────────────────────────────
⚡ Gradient  ·  LLMs & Transformer Architecture

Attention Is All You Need came out in 2017. Seven years
later, we're still finding things hidden in plain sight.
The positional encodings? Basically a Fourier transform.
The residual stream? A memory bus. Read papers like
you're debugging code.

#Transformers #MachineLearning #DeepLearning

█▇█▆  Deep dive + breakthrough — hidden structure revealed

───────────────────────────────────────────────────────────

SESSION SUMMARY
───────────────────────────────────────────────────────────
Posts displayed: 2

Cumulative profile:    Session blend:
▇▇█▆                   Heavy on insight and breakthrough,
L E B M                moderate mood lift, strong learning

Strongest dimension:   Breakthrough (avg ▇)
Weakest dimension:     Mood (avg ▆)

Run /socialfeed to generate more posts