analyze-feed

Analyze LinkedIn Feed

Read the SQLite database of scraped LinkedIn posts and generate intelligent analysis — topic clustering, engagement scoring, trend detection, and content gap identification.

Data Location

The SQLite database is at ${CLAUDE_PLUGIN_ROOT}/scripts/data/feeds.db. Use Bash to query it.

Analysis Process

Step 1 — Load Recent Data

Query the database for posts from the last 7 days (or the user's requested window):

SELECT author_name, content, likes, comments, engagement, collected_at, post_type
FROM posts
WHERE collected_at >= datetime('now', '-7 days')
ORDER BY collected_at DESC;

Also load the user's topic configuration:

SELECT value FROM config WHERE key = 'topic_clusters';

If no topic clusters are configured, use these defaults: Founders & Startups, Operations & Productivity, Sales & Revenue, Work & Culture.

Step 2 — Topic Clustering

Read each post and assign it to one or more topic clusters. Do NOT use keyword matching — actually read and understand the content. A post about "our fundraising journey nearly destroyed my marriage" belongs in Founders AND Work & Culture, not just whichever one has a keyword match.

For each cluster, track: post count, total engagement (likes + comments), top 3 posts by engagement, and notable authors.

Step 3 — Trend Detection

Compare this week's topic distribution to the previous week:

SELECT author_name, content, likes, comments, engagement, collected_at
FROM posts
WHERE collected_at >= datetime('now', '-14 days')
AND collected_at < datetime('now', '-7 days')
ORDER BY collected_at DESC;

Identify topics that are growing, declining, or newly emerging. Flag any sudden spikes.

Step 4 — Content Gap Analysis

Identify topics where the network is active but the user hasn't posted. Cross-reference topic clusters against the user's own posts (where author_name matches the user's LinkedIn name from config).

Step 5 — Engagement Patterns

Identify which types of posts get the most engagement: long-form vs. short, question posts vs. statements, posts with images vs. text-only, time of day patterns.

Output Format

Present findings conversationally — not as a raw data dump. Lead with the most interesting insight, then support with data. Use specific examples from actual posts (cite the author and a brief content snippet). End with 2-3 actionable content opportunities based on the gaps and trends found.

Database Schema Reference

See references/database-schema.md for the full SQLite schema including all tables and relationships.