News Signal Outreach

News Signal Outreach

The long-tail composite. Scripts ingest the news and find contacts; the agent extracts entities, evaluates ICP fit, finds the relevance angle, and drafts. Many requests are evaluate-only — stop after the angle and don't draft/send unless asked.

When to use

• User shares ANY URL/post/article: "is this relevant / a fit / should we reach out?"
• "Came across", "saw this", "found this article" + a company/person.
• Route here only what the four structured composites (funding/hiring/leadership/champion) don't own.

How it works (scripts + agent)

1. Ingest — ingest_news.py --url <link> fetches + cleans the page (or --text-file for pasted text). LinkedIn/tweet/gated sources flag partial_content -> use web-automation + cookie in a flow. (agent) extract companies, people, event type.
2. (agent) Evaluate ICP fit of each entity; for person-level news, resolve their company (web search / Apollo).
3. (agent) Find the relevance angle — how the news shifts priorities / opens a window for the product. For evaluate-only requests, stop here and return fit + angle.
4. Find people — find_contacts.py (Apollo by domain + persona titles; --enrich --verify; degrades to pattern-guess).
5. (agent) Draft sequences with email-drafting using the news as the hook — keep personalization deep (Tier 3); these are opportunistic and un-templated.
6. Hand off to cold-email-outreach / linkedin-outreach; request launch approval.

How to run

# 1 — ingest the news
python3 ${SKILL_DIR}/scripts/ingest_news.py \
  --url "https://example.com/acme-acquires-foo" --output ${WORKSPACE}/news.json

# 4 — find people (only if proceeding to outreach)
python3 ${SKILL_DIR}/scripts/find_contacts.py \
  --domains ${WORKSPACE}/qualified.json --titles "VP Ops,Head of IT" \
  --per-company 3 --enrich --verify --output ${WORKSPACE}/contacts.json

Outputs

• news.json (cleaned content), the agent's icp-qualified-targets + relevance-angles, contacts.json, drafted sequences. Evaluate-only: just the qualified targets + angle.

Credentials / env

• env.required: none. Ingest is keyless, reasoning/drafting is the agent, and the launch step hands off to cold-email-outreach (CSV export with no key).
• env.optional (all degrade): APOLLO_API_KEY — if set → people-finding / company resolution; else → keyless serp + pattern-guess. MILLIONVERIFIER_API_KEY — if set → verify; else → local syntax/dedup (bounce risk). DROPCONTACT_API_KEY — email-finding fallback. Send/launch — if a send key (LEMLIST_API_KEY / INSTANTLY_API_KEY / SENDGRID_API_KEY / RESEND_API_KEY, or PHANTOMBUSTER_API_KEY for LinkedIn) is set → launch the sequence; else → export a CSV to send manually (the keyless default).

Notes & edge cases

• Many requests are evaluate-only — stop after the angle; don't draft/send unless asked.
• LinkedIn/tweet sources need web-automation (JS/anti-bot); some posts are gated and yield partial content (degrade to LLM-from-snippet).
• Keep personalization deep — the best of these are un-templated.