Trending Ad Hook Spotter

Trending Ad Hook Spotter

Composite: multi-platform social scan → trend detection + scoring → hook translation → report. The HN scan is a keyless deterministic script; X/Reddit/LinkedIn scanning is agent-driven web search (with an optional Apify depth fallback). Trend clustering, scoring, urgency tiering, and hook writing are the agent's reasoning — no LLM call in a script. Recency is everything: bound every scan to the last ~7 days.

When to use

• "What's trending in our space that we could run ads about?"
• "Find viral hooks for our paid campaigns" / "Newsjacking opportunities for <client>."
• "What should we be running ads about this week?"
• Good cron candidate for a weekly "what to run this week" digest.

How to run

1 — Hacker News scan (keyless, recency-bounded)

python3 ${SKILL_DIR}/scripts/hn_trends.py --keywords "rpa,workflow automation,ai agents" \
  --days 7 --output ${WORKSPACE}/hn.json
# Front-page sweep too:
python3 ${SKILL_DIR}/scripts/hn_trends.py --keywords "observability" --tags front_page --days 3

Returns items ranked by an engagement-velocity proxy (points + comments per hour since posting), so old high-engagement threads don't masquerade as trends.

2 — X / Reddit / LinkedIn scan (you, the agent — keyless web search)

Drive via web search bounded to the past ~7 days (Robomotion Proxy). SERP path: if DATAFORSEO_LOGIN/DATAFORSEO_PASSWORD (or SERPER_API_KEY) is set → structured SERP API; if not → the agent's keyless web search (default).

• X — <keyword> (viral OR thread) site:x.com, competitor-momentum (launches/outages/ funding) and frustration queries.
• Reddit — site:reddit.com topic/hot threads; flag high upvote/comment ratios, buying-intent and switching threads.
• LinkedIn — site:linkedin.com/posts / site:linkedin.com/pulse for KOL + competitor posts.

For depth beyond search (X/Reddit at volume): if APIFY_API_TOKEN is set → use the Apify runner (better depth/velocity); if not → degrade to site: web search (default). Apify runner:

python3 ${SKILL_DIR}/scripts/apify_run.py --actor "trudax~reddit-scraper" \
  --input '{"searches":["ai agents"],"sort":"hot","maxItems":60}' --output ${WORKSPACE}/reddit.json

Without APIFY_API_TOKEN, degrade X/Reddit to site: search (lower velocity precision — note it). HN always works key-free. LinkedIn engagement depth: if PHANTOMBUSTER_API_KEY (+ a LinkedIn session cookie) is set → PhantomBuster path; if not → site:linkedin.com web search (default).

3 — Detect + score trends (you, the agent)

Cluster signals into trends (cross-platform within 7 days, OR one post with ~10×-norm engagement, OR a breaking event). Normalize velocity for account size/age so a small viral post isn't drowned by a big account's baseline. Score each trend: recency 25 / velocity 25 / cross-platform 20 / ICP relevance 20 / product fit 10 → urgency tier (Run Today 90+, This Week 70–89, Worth Testing 50–69). If content_velocity is "weekly", down-weight 24–48h trends the team can't act on in time.

4 — Hook translation (you)

For each trend ≥50: a 2-sentence summary, ad-opportunity rationale, 3 hook variants (newsjack / contrarian / practical), recommended format + platform, and a time window (gated by content_velocity).

5 — Render

Write trending-hooks-<YYYY-MM-DD>.md to ${WORKSPACE} and attach to the Agent Teams channel. For a recurring digest, schedule via the platform cron and dedup seen trends across runs (SUPABASE_* if persisting).

Outputs

trending-hooks-<YYYY-MM-DD>.md — trends grouped by urgency tier, each with a 2-sentence summary, engagement signal, time window, and 3 hook variants + recommended format/platform; a trend-velocity dashboard (platform × score × window) and a competitor-trend-involvement table.

Credentials / env

• Required: none — scanning runs on web search + the keyless HN API. Trend detection, scoring, and hook writing are the agent's reasoning (no LLM key in scripts).
• Optional (each with a keyless default fallback):
  • APIFY_API_TOKEN — if set → Apify Reddit/X actor for depth; else → site: web search + the keyless HN API (default).
  • PHANTOMBUSTER_API_KEY (+ LinkedIn session cookie) — if set → PhantomBuster LinkedIn engagement depth; else → site:linkedin.com web search (default).
  • SUPABASE_URL/SUPABASE_KEY — if set → dedup seen trends across cron runs; else → single-run output (default).
  • DATAFORSEO_LOGIN/DATAFORSEO_PASSWORD (or SERPER_API_KEY) — if set → structured SERP API for the social-scan queries; else → the agent's keyless web search (default).
  • HTTPS_PROXY — Robomotion Proxy for SERP scans.

Notes & edge cases

• Recency is everything — bound every scan to ~7 days; an old high-engagement thread is not a trend.
• Normalize engagement velocity for account size/age.
• Without APIFY_API_TOKEN, X/Reddit degrade to site: search (note lower precision); HN always works key-free.
• Throttle + proxy all SERP scans.