X/Twitter Scanner

What This Does

Uses xAI's Grok API with the built-in x_search tool to scan X/Twitter for recent posts from AI-focused accounts. Grok has native access to X data, so one API call handles search, filtering, and summarization. No separate Twitter API credentials needed.

The scan produces a structured digest tagging launches, tools worth trying, learning material, build ideas, and content angles — with a coverage check so your must-watch accounts never silently drop out.

Setup

The script reads the xAI API key from the environment first, then falls back to a .env file (it walks up the directory tree to find one). Either works:

export XAI_API_KEY=your_xai_api_key_here
# or add to a .env file at your project root:
# XAI_API_KEY=your_xai_api_key_here

Get an API key from https://x.ai/api. No external Python packages are needed — the script uses the standard library only.

Configuring Who You Track — watchlist.json

Accounts live in watchlist.json next to the scripts/ folder. Edit that file, not the script, to track whoever you want. It has two parts:

• must_surface_handles — accounts the scan always tries to cover. If one is missed on the first pass, the scanner automatically re-scans just those handles and merges them in. A coverage footer reports how many were surfaced.
• groups — everyone else, organized into named buckets (AI labs, builders, news, etc.). Each group is scanned together, and grouped scans kick in as a fallback if a broad pass comes back thin. Handles can be plain strings or { "handle": "...", "note": "..." } objects — notes are just for your own reference.

If watchlist.json is missing or invalid, the script falls back to a small set of baked-in default handles so it still runs.

How It Works

The script calls https://api.x.ai/v1/responses with the x_search tool enabled. Grok searches X in real-time and returns a summarized digest. The scanner adds reliability on top:

1. Broad pass over all watchlist accounts.
2. If the result is thin or misses required accounts, it escalates — scanning curated groups separately, then falling back to a must-surface-only scan.
3. It re-scans any still-missing required handles and merges them in.
4. It appends a deterministic coverage check and exits non-zero if nothing usable came back.

Digest Structure

Every account scan returns these sections:

• FLAGGED — high-signal posts: launches, real revenue/growth numbers, contrarian takes, adoption-worthy workflows, or viral engagement.
• TRY THIS — concrete tools, models, APIs, or integrations you can go use right now. Publicly callable models, free-tier or open-source releases, new API capabilities, or replicable "here's exactly how" posts. Not research papers, waitlists, or "coming soon."
• LEARN NEXT — educational content worth studying (frameworks, tutorials, becoming-table-stakes skills).
• BUILD IDEAS — concrete product/automation/agent ideas inspired by the scan.
• CONTENT ANGLES — angles usable across LinkedIn, X, newsletter, or short-form video.
• TOP FEED — remaining notable posts, ranked by relevance.
• COVERAGE CHECK — how many must-surface handles were surfaced, and which were missed.

How to Use

Quick scan (default — last 48 hours)

python <skill-directory>/scripts/scan_x.py

Custom time window

python <skill-directory>/scripts/scan_x.py --hours 6

Specific accounts only (skips the fallback machinery — scans exactly what you pass)

python <skill-directory>/scripts/scan_x.py --handles karpathy sama AnthropicAI

Freeform X search

python <skill-directory>/scripts/scan_x.py --query "Claude Code update"

Summary only (no JSON wrapper)

python <skill-directory>/scripts/scan_x.py --summary-only

Output is JSON to stdout with a summary field containing Grok's digest (or just the digest text with --summary-only). The script exits with code 2 if the scan returned nothing usable.

Default Workflow

1. Run scan_x.py to pull recent posts via the xAI Grok API.
2. If the script fails (key expired, rate limited), fall back to a web search with queries like "AI news today", "Claude update", "OpenAI announcement".
3. Present the digest to the user, leading with FLAGGED and TRY THIS.
4. STOP here. Don't generate downstream content or push to any external tool unless the user explicitly asks.

Automated Scanning

scripts/scheduled_scan.sh writes a timestamped digest to a digests/ folder next to the script — wire it into cron for hands-off daily scans:

# Scan twice a day at 8am and 8pm
0 8,20 * * * /path/to/skills/x-scanner/scripts/scheduled_scan.sh