Competitive Feature Matrix Builder

What this skill does

Crawls competitor product pages and pricing pages, extracts feature claims, and builds a feature-by-vendor matrix. Each cell cites the source URL where the claim was found. Flags features your product is missing.

Inputs

• vendors: array of {name, urls: [pricing_url, features_url, docs_url], own?: true}. Mark exactly one as own.
• features: optional array of feature names to seed; otherwise discovered.
• Optional out_path: defaults to matrix.md.

Steps

1. For each vendor URL, fetch the page using curl -L --user-agent 'AmitteResearcher/1.0' --max-time 15. Save HTML to /tmp/<vendor>-<page>.html.
2. Strip nav and footer with a small reader: keep only the <main> or <article> section if present.
3. Extract feature claims: bullet items inside feature/pricing tables, H2/H3 plus their following sentence, and check-mark icons next to feature names.
4. Normalize feature names: lowercase, strip punctuation, lemmatize (e.g., "logins" -> "login").
5. If features was provided, intersect; else, take the union of all observed features and dedupe by similarity (Jaccard 0.6).
6. For each (feature, vendor) cell, mark present (claim found), absent (no mention on the relevant pages), or unclear.
7. For each cell, attach the source URL and a short quoted snippet (max 120 chars).
8. Identify gaps: features marked present for at least 2 competitors but absent for the own vendor.
9. Render a markdown table with vendors as columns and features as rows. Append a "Gaps" section listing the gaps with sources.
10. Save the matrix.

Output

matrix.md with the feature table, a Gaps section, and a "Methodology" footer naming the URLs crawled and the date. Plus a matrix.csv for spreadsheet import. Stdout prints the gap count.

Verification

Sample three present cells and visit the cited URL to confirm the snippet matches. For three absent cells, search the vendor's site (site:vendor.com <feature>) to confirm the feature isn't on a less-prominent page; promote to unclear if any hit is found. Re-crawl in a week and diff: vendor pages change, and a stale matrix is misleading.

Edge cases

• Vendor pages that gate behind login: cannot crawl; mark cells unverified and rely on docs or G2/Capterra reviews.
• Marketing pages that conflate roadmap with shipped features: any "coming soon" or "beta" badge demotes the cell to partial.
• Feature naming differences ("SSO" vs "Single Sign-On"): rely on the lemma normalization plus a small alias table.
• The own vendor's own features list is incomplete: source-of-truth for "own" should be the actual product, not a pricing page; surface a self-audit recommendation.