agentleFS
Othmane-KhadriOthmane-Khadri/YALC-the-GTM-operating-system/.claude/skills/prospect-discovery-pipeline/SKILL.md

prospect-discovery-pipeline

Skill298 starsChanged 6 months agoMITView on GitHub

Use when a teammate wants a full discovery-to-outreach pipeline anchored on existing clients. Triggers include "find prospects like [client]", "build a target list like [domain]", "lookalike discovery for [client]", "discovery to outreach for [criteria]", "10 companies similar to [X] with a CMO", or any multi-step request combining lookalike search + decision-maker identification + signal enrichment + LinkedIn variant drafting.

  • Reads credentials

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 2 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Always follow this order. Quote credit cost before each phase.

    line 20
  2. Very specific:

    If people_enrich is needed: +30 Crustdata credits.

    line 100
  3. Distinctive:

    Never push to Notion / activate Unipile campaign without explicit user approval at the second checkpoint.

    line 108
  4. Very specific:

    Crustdata domain mismatch: search by company name as fallback when domain returns 0.

    line 119

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Prospect Discovery Pipeline

End-to-end pipeline: PredictLeads lookalikes → ICP filter → Crustdata CMO finder → multi-signal enrichment → 2 LinkedIn variants drafted with per-lead personalization. Pauses for user review before any expensive operation. Always quotes credit cost up front.

When to use

  • Building a target account list anchored on 1–2 known clients
  • Generating a campaign-ready batch (10–25 leads with full signal context)
  • Producing 2 A/B-testable LinkedIn message variants tied to actual signal data per lead

Don't use when: ad-hoc lookup of one company (use predictleads-signals); just lookalike domains without contacts (use predictleads-lookalikes); enriching a list you already have qualified leads for (use signals:enrich --result-set directly).

The 5-phase flow

Always follow this order. Quote credit cost before each phase.

Phase 1 — Discovery (2 PL credits for 2 anchors)

npx tsx src/cli/index.ts signals:similar --domain anchor1.com --limit 50
npx tsx src/cli/index.ts signals:similar --domain anchor2.com --limit 50

Merge into a candidate pool, dedupe by domain. Expect 30–80 unique candidates per pair.

Phase 2 — ICP filter (FREE, pause for user review)

Hand-filter the pool against the user's ICP criteria:

  • Employee count (use Crustdata company_identify — FREE — only when judgement uncertain)
  • Industry vertical (back-office SaaS, commerce infra, HR-tech, etc.)
  • HQ region
  • Marketing maturity proxies (visible content investment)

STOP and present the 10 finalists to the user before spending more credits. Surface any obvious gaps or weak fits. Wait for explicit approval.

Phase 3 — CMO finder (3 Crustdata credits, batch)

Single batch search across all 10 companies:

filters = {
  op: 'and',
  conditions: [
    { column: 'current_employers.company_website_domain', type: 'in', value: ['10 domains'] },
    { column: 'current_employers.title', type: '[.]', value: 'Marketing' },
    { column: 'current_employers.seniority_level', type: 'in', value: ['CXO', 'Vice President', 'Director'] },
  ],
}
limit: 50

Pick 1 marketing leader per company (prefer CMO > VP > Head > Director).

Common gotcha: some companies' websites are stored in Crustdata as ATS or marketing domains (e.g., hubs.li for Shopware), not their actual .com. If a company returns 0 hits, do a fallback search by current_employers.name substring.

Skip people_enrich unless the campaign needs emails (LinkedIn-only campaigns don't). Saves ~30 credits.

Phase 4 — Multi-signal enrichment (40 PL credits for 10 finalists)

for d in domain1.com domain2.com ...; do
  npx tsx src/cli/index.ts signals:fetch --domain "$d"
done

Or use the bulk shortcut if leads already in a result set:

npx tsx src/cli/index.ts signals:enrich --result-set <id>

Phase 5 — Hydrate templates + draft 2 variants (FREE)

Pick the single most outreach-relevant signal per company (most recent news > recent financing > recent job_opening). Build a personalization_natural line per lead that:

  • Never says "I saw your [signal]" (per outbound rules)
  • Embeds the signal as context for a category insight
  • Stays ≤18 words per sentence
  • Has no dashes, no I openers, says Hello, ends with a specific CTA

Draft both variants with different angles (e.g., results-led case study vs. category-shift narrative). Save the full draft to 00_Inbox/predictleads-discovery-{date}.md.

Do not push to Notion or activate the campaign without explicit user approval.

Total cost (typical)

PhaseCredits
1. Lookalikes (2 anchors)2 PL
2. ICP filter0
3. CMO batch search3 Crustdata
4. Multi-signal enrichment (10 companies × 4 types)40 PL
5. Template hydration0
Total~45 credits (42 PL + 3 Crustdata)

If people_enrich is needed: +30 Crustdata credits.

Verification checkpoints

The pipeline pauses at:

  1. End of Phase 2 — present 10 finalists, wait for "approved"
  2. End of Phase 5 — present hydrated drafts, wait for "approved"

Never push to Notion / activate Unipile campaign without explicit user approval at the second checkpoint.

Output artifacts

  • SQLite: company_signals rows for the 10 finalists
  • File: 00_Inbox/predictleads-discovery-{date}.md with the 2 hydrated variants
  • Optional: HTML dashboard via predictleads-dashboard skill

Common pitfalls

  • Megacaps in the lookalike pool: PredictLeads returns SAP/Microsoft/Oracle for B2B SaaS seeds. Filter manually before Phase 3.
  • Crustdata domain mismatch: search by company name as fallback when domain returns 0.
  • Personalization that flag-waves: "I saw your funding round" violates outbound rules. Reframe as category context.

Required env

PREDICTLEADS_API_KEY, PREDICTLEADS_API_TOKEN, CRUSTDATA_API_KEY in ~/.gtm-os/.env. See TEAM_SETUP.md.

Related skills

  • predictleads-signals — single-company ad-hoc
  • predictleads-lookalikes — discovery only, no outreach
  • predictleads-dashboard — HTML viz of enriched signals
  • unipile-campaign — what runs the actual outreach after this skill drafts the variants

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.