Battlecard Generator

Battlecard Generator

Composite: deterministic scripts collect public competitor signal (pages, reviews, ads, pricing); you, the agent, do all scoring and write the battlecard. Scoped to ONE competitor — breadth dilutes the card.

When to use

• "Build a battlecard against [competitor]." / "Help me win deals against [competitor]."
• "We keep losing to [competitor] — why? What weaknesses can we exploit?"
• Prepping a sales team for competitive deals or entering a market with an incumbent.

How to run

1. Capture competitor pages (keyless)

python3 ${SKILL_DIR}/scripts/fetch_pages.py \
  --url https://competitor.com https://competitor.com/pricing \
        https://competitor.com/about https://competitor.com/product \
  --output ${WORKSPACE}/competitor_pages.json

Returns per page: title, meta description, headings, paragraphs, list items, candidate_claims (hero/value-prop lines), links, and stats.likely_js_rendered.

If a page has likely_js_rendered: true (or error), re-fetch it rendered:

npx playwright install chromium   # first run only
node ${SKILL_DIR}/scripts/render_page.mjs \
  --url https://competitor.com/pricing --output ${WORKSPACE}/competitor_pricing_rendered.json

2. Mine reviews (G2 / Capterra — JS + anti-bot, use the renderer)

Find the review URL with your own web search ("[competitor]" site:g2.com / site:capterra.com), then render it:

node ${SKILL_DIR}/scripts/render_page.mjs \
  --url "https://www.g2.com/products/<competitor>/reviews" \
  --selector "[itemprop='review'], .paper--white" \
  --output ${WORKSPACE}/g2_reviews.json

selected.text holds the review blocks. Pull top praised features (their moat) and top complaints (your attack angles). If blocked, fall back to APIFY_API_TOKEN (see Notes).

3. Ads & social signal

• Ads: render Meta Ad Library / Google Ads Transparency Center result pages with render_page.mjs (use --wait 6000, they hydrate slowly).
• Social/community: use your own web search for "[competitor]" site:reddit.com and community frustrations; render specific threads if needed.

4. Synthesize the battlecard (you, the agent — no script)

Read all the collected JSON and write an opinionated markdown battlecard with: 30-second quick reference, competitor overview, positioning traps, landmine questions, objection handlers, honest feature comparison (Them / Us / Net), "their customers say" (real review quotes only), pricing comparison + attack angle, win/loss themes, and ready-to-paste email/chat responses. Stamp date + a confidence rating keyed to data freshness. "Where We Lose" must include mitigation, not just admission.

Save to ${WORKSPACE}/battlecard-[competitor]-[YYYY-MM-DD].md and post as a channel attachment.

Outputs

• ${WORKSPACE}/competitor_pages.json, g2_reviews.json, etc. — raw collected signal.
• ${WORKSPACE}/battlecard-[competitor]-[date].md — the rep-facing battlecard (your synthesis).

Credentials / env

• Required: none — fetch_pages.py and render_page.mjs are keyless; synthesis is the agent (platform-default model).
• Optional: APIFY_API_TOKEN — deeper G2/Capterra/Reddit scraping when the renderer is blocked. Degrades to keyless fetch + browser without it. DATAFORSEO_LOGIN/ DATAFORSEO_PASSWORD or SERPER_API_KEY — if set, the review/ad/social discovery searches use a paid SERP; if not, the agent's own web search (the default) does them.

Notes & edge cases

• G2/Capterra and ad libraries are hostile, JS-heavy targets — always use render_page.mjs (a real headless browser), not the urllib fetcher. Escalate to Apify only on a hard block.
• Apify degrade path (when set): curl -s "https://api.apify.com/v2/acts/<actor>/run-sync-get-dataset-items?token=$APIFY_API_TOKEN" -H 'Content-Type: application/json' -d '{...}'
• Keep it to ONE competitor. Output must be opinionated, not a neutral feature grid.
• Never invent review quotes or ad copy — if a section can't be retrieved, mark it "[NEEDS VERIFICATION]" and still ship the card from website + pricing.
• Stamp data freshness and confidence; reviews and ads age fast.