hyreflow
Get started
← all skills
skill

Layoff signal → poach displaced talent (+ BD)

Skill — Layoff signal → poach displaced talent (+ BD)

Use when: the trigger is "a company is cutting staff." Either a broad sweep ("who laid off in the last 7 days?") or a targeted employer ("Microsoft just announced layoffs — get me the displaced engineers"). A layoff = a pool of available candidates to poach and a BD opening with the survivors.

Powered by the layoffsignal Hyreflow Native (free public news RSS) → fan into the standard sourcing pipeline. The native is free (0 credits), no BYOK (natives are exempt from the waterfall).

Where this runs (Model A)

The extraction + dedupe + qualification is reasoning done by the customer's own Claude — free, no hosted AI model call. The feed pull is free too — credits are spent on the downstream enrichment of passes only.

Inputs

  • Mode: broad (whole market) or targeted (one or more named employers).
  • when window (default 7d; 30d for a named company).
  • The client's ICP.md (for the qualify gate) — per-client data, not in this skill.

Step 1 — Sweep the feeds (layoffsignal, native — credits)

from lib.layoffsignal import LayoffSignal
ls = LayoffSignal()
items = ls.layoff_news(when="7d")                       # broad sweep
# or targeted:  ls.layoff_news(company="Microsoft", when="30d")
# or full sweep stream: ls.iter_layoff_news(when="7d", companies=["Microsoft","Meta"])
# clean DIRECT links for curated outlets: ls.site_feeds()

Each item → {title, link, source, published, summary}. Links from google_news/layoff_news are Google News redirector URLs; site_feed() links are clean/direct.

Step 2 — Resolve + fetch article bodies (firecrawl)

For each item, get the real article text. firecrawl follows the redirector and returns the body (more reliable than resolve_link() for the encoded Google News links). Skip paywalled fetches; keep the headline+source as the minimal signal.

Step 3 — Extract structured rows (Claude, free — Model A)

From each article, the agent extracts: {company, headcount, pct, date, location, function/dept (if stated), source_url, source_publisher}. Articles are often vague ("hundreds") — record the raw phrasing; don't fabricate a precise number.

Step 4 — Dedupe + normalize (code, free)

Collapse multiple articles about the same event: key on (normalized company, ~date). Keep the most specific headcount and the best source. Output one row per layoff event.

Step 5 — Resolve company → source affected roles

  • Resolve company → domain (builtwith.company_to_url, or known domain).
  • Qualify the company vs ICP.md if this is BD (recipes/qualify-against-icp.md).
  • Source the displaced people by function:
    • Engineering/IT → github (search users by company/location/language; find_user_emails).
    • Everyone else → apollo / aiark people-search at that company (recently-departed if available), walking the people_search provider order.

Step 6 — Enrich → verify → outreach

🎯 Channel decision — ASK THE USER before enriching. This is candidate acquisition, so the default is LinkedIn + personal email ONLY — never work email (you're recruiting them away from that employer; their work inbox is wrong and often dead post-layoff). Prompt: "Personal email + LinkedIn only (recruiting default), or also include work email?" The answer picks the waterfall:

  • personal (default) → personal_email order: leadmagic(personal) → fullenrich(personal) → prospeo → icypeas (+ GitHub find_user_emails for engineers).
  • work (only if user opts in, e.g. BD) → email_enrichment (work) order.

LinkedIn is the first/primary touchpoint — we almost always already have it from people-search, so it needs no enrichment. Personal email is the second channel when found. SourceWhale is multichannel → one cadence fires a LinkedIn touch + an email touch on the same candidate (utilize both).

Standard tail: personal-email/phone waterfall (per the channel decision above) → validate (enrichley) → sequence (sourcewhale for recruiting / lemlist / instantly) → land in the client's ATS. Approval-gate the first send + any bulk enrichment (one-record pilot first).

⚠️ SourceWhale prerequisite — no create-campaign API. Before pushing candidates, tell the user they must create the campaign + message steps manually in the SourceWhale app, and wait for confirmation; then resolve its id via list_campaigns. Never attempt to create a campaign via API. (See tools/sourcewhale.md.)

Pipeline

layoffsignal (native) → fetch (firecrawl) → extract (Claude) → dedupe
   → [qualify vs ICP, if BD] → source (github | apollo/aiark) → enrich → verify → sequence → ATS

Guardrails

  • ToS / attribution. The native surfaces facts from public reporting — attribute the originating publisher; never republish article text or resell a curated dataset. Signal/trigger use only.
  • Don't fabricate headcounts. Keep the source's own wording when imprecise.
  • Recall caveat. An automated sweep won't catch everything a manual tracker does — tune the query terms (LAYOFF_TERMS) and the curated DEFAULT_SITES over time; widen when for thoroughness.
  • Spend. The native costs credits per pull; enrichment costs credits per record — pilot small.