X Bookmarks v2

Turn X/Twitter bookmarks from a graveyard of good intentions into actionable work.

Core philosophy: Don't just summarize — propose actions the agent can execute.

Data Source Selection

This skill supports two backends. Pick the first one that works:

1. bird CLI (preferred if available)

• Fast, no API key needed, uses browser cookies
• Install: npm install -g bird-cli
• Test: bird whoami — if this prints a username, you're good

2. X API v2 (fallback)

• Works without bird CLI
• Requires an X Developer account + OAuth 2.0 app
• Setup: see references/auth-setup.md → "X API Setup"

Auto-detection logic

1. Check if `bird` command exists → try `bird whoami`
2. If bird works → use bird CLI path
3. If not → check for X API tokens (~/.config/x-bookmarks/tokens.json)
4. If tokens exist → use X API path (auto-refresh)
5. If neither → guide user through setup (offer both options)

Fetching Bookmarks

Via bird CLI

# Latest 20 bookmarks (default)
bird bookmarks --json

# Specific count
bird bookmarks -n 50 --json

# All bookmarks (paginated)
bird bookmarks --all --json

# With thread context
bird bookmarks --include-parent --thread-meta --json

# With Chrome cookie auth
bird --chrome-profile "Default" bookmarks --json

# With manual tokens
bird --auth-token "$AUTH_TOKEN" --ct0 "$CT0" bookmarks --json

If user has a .env.bird file or env vars AUTH_TOKEN/CT0, source them first: source .env.bird

Via X API v2

# First-time setup (opens browser for OAuth)
python3 scripts/x_api_auth.py --client-id "YOUR_CLIENT_ID" --client-secret "YOUR_SECRET"

# Fetch bookmarks (auto-refreshes token)
python3 scripts/fetch_bookmarks_api.py -n 20

# All bookmarks
python3 scripts/fetch_bookmarks_api.py --all

# Since a specific tweet
python3 scripts/fetch_bookmarks_api.py --since-id "1234567890"

# Pretty print
python3 scripts/fetch_bookmarks_api.py -n 50 --pretty

The API script outputs the same JSON format as bird CLI, so all downstream workflows work identically.

Token management is automatic: tokens are stored in ~/.config/x-bookmarks/tokens.json and refreshed via the saved refresh_token. If refresh fails, the agent should guide the user to re-run x_api_auth.py.

Environment variable override

If the user already has a Bearer token (e.g., from another tool), they can skip the OAuth dance:

X_API_BEARER_TOKEN="your_token" python3 scripts/fetch_bookmarks_api.py -n 20

JSON Output Format (both backends)

Each bookmark returns:

{
  "id": "tweet_id",
  "text": "tweet content",
  "createdAt": "2026-02-11T01:00:06.000Z",
  "replyCount": 46,
  "retweetCount": 60,
  "likeCount": 801,
  "bookmarkCount": 12,
  "viewCount": 50000,
  "author": { "username": "handle", "name": "Display Name" },
  "media": [{ "type": "photo|video", "url": "..." }],
  "quotedTweet": { "id": "..." }
}

Core Workflows

1. Action-First Digest (Primary Use Case)

The key differentiator: don't just summarize, propose actions the agent can execute.

• Fetch bookmarks (bird or API, auto-detected)
• Parse and categorize by topic (auto-detect: crypto, AI, marketing, tools, personal, etc.)
• For EACH category, propose specific actions:
• Tool/repo bookmarks → "I can test this, set it up, or analyze the code"
• Strategy/advice bookmarks → "Here are the actionable steps extracted — want me to implement any?"
• News/trends → "This connects to [user's work]. Here's the angle for content"
• Content ideas → "This would make a great tweet/video in your voice. Here's a draft"
• Questions/discussions → "I can research this deeper and give you a summary"
• Flag stale bookmarks (>2 weeks old) — "Use it or lose it"
• Deliver categorized digest with actions

📂 CATEGORY (count)
• Bookmark summary (@author)
→ 🤖 I CAN: [specific action the agent can take]

2. Scheduled Digest (Cron)

Set up a recurring bookmark check. Suggest this cron config to the user:

Schedule: daily or weekly
Payload: "Check my X bookmarks for new saves since last check.
  Fetch bookmarks, compare against last digest, summarize only NEW ones.
  Categorize and propose actions. Deliver to me."

Track state by saving the most recent bookmark ID processed. Store in workspace: memory/bookmark-state.json → { "lastSeenId": "...", "lastDigestAt": "..." }

3. Content Recycling

When user asks for content ideas from bookmarks:

• Fetch recent bookmarks
• Identify high-engagement tweets (>500 likes) with frameworks, tips, or insights
• Rewrite key ideas in the user's voice (if voice data available)
• Suggest posting times based on the bookmark's original engagement

4. Pattern Detection

When user has enough bookmark history:

• Fetch all bookmarks (--all)
• Cluster by topic/keywords
• Report: "You've bookmarked N tweets about [topic]. Want me to go deeper?"
• Suggest: research reports, content series, or tools based on patterns

5. Bookmark Cleanup

For stale bookmarks:

• Identify bookmarks older than a threshold (default: 30 days)
• For each: extract the TL;DR and one actionable takeaway
• Present: "Apply it today or clear it"
• User can unbookmark via: bird unbookmark (bird only)

Error Handling

Tips

• Start with -n 20 for quick digests, --all for deep analysis
• bird: Use --include-parent for thread context on replies
• API: includes bookmarkCount and viewCount (bird may not)
• Bookmark folders supported via bird --folder-id
• Both backends output identical JSON — workflows are backend-agnostic