The Problem

Workflow Steps

1. 1
   Define a JSON config file with each competitor's name and the URLs to monitor: pricing, features, changelog, and careers pages
2. 2
   Scrape Agent: use Firecrawl (preferred for JS-rendered pages) or Playwright to scrape each page and convert to markdown — store as {competitor, page_type, content, scraped_at}
3. 3
   Memory Retriever: query ChromaDB for the previous week's snapshot for each {competitor, page_type} combination using metadata filters
4. 4
   Diff Agent: pass both the current and previous content to Claude with a semantic diff prompt — return JSON array of detected changes with change_type, old_value, new_value, and significance rating; return empty array if no meaningful changes detected
5. 5
   Signal Classifier: for each detected change, classify as: pricing_change / feature_launch / positioning_shift / hiring_signal / deprecation / risk_signal
6. 6
   Analyst Agent: interpret the classified changes strategically — what does this signal about their direction, which changes threaten your positioning, which open an opportunity, what should your team monitor next
7. 7
   Memory Updater: store the new snapshot using a deterministic document ID per {competitor, page_type} combination (e.g. "competitorx_pricing"). Use ChromaDB's collection.update() if the document ID exists, or collection.add() on first run — this ensures each competitor-page combination always has exactly one snapshot and the previous version is replaced cleanly

Prompt Example