client-sourcing

client-sourcing

A Claude skill that runs a value-first outbound pipeline for a solo or fractional consulting practice — turn signals into qualified, well-pitched conversations without turning into a spam cannon.

It encodes a simple thesis: when your reputation is the brand, the win condition is reply rate and relationship quality, not volume. So every touch is personalized, grounded in real proof, and leads with an insight you give before you ever ask for anything.

Built as a Claude skill (works in Claude Desktop / Cowork and Claude Code). Drop in your own profile and it becomes your sourcing co-pilot.

Preview

A scored sweep, rendered two ways — the weekly digest, and the same leads imported as Salesforce Leads:

Both are live output formats in templates/; open the .html versions to reuse them. Data is a fictional worked example.

What it does

• Sources leading signals, not lagging ones. Funding rounds, exec moves, launches, and freezes — upstream of the job post everyone else is answering.
• Qualifies ruthlessly. A four-axis scorer (fit · budget · timing · edge) with hard ICP gates, so you don't spend effort on the unfundable or the off-target.
• Pitches with receipts. Templates that lead with a specific insight and one clear ask — no "let me help you" availability-speak.
• Ships in your format. One scored sweep → a plain-text email, a Slack brief, a pretty HTML digest, or a Salesforce-importable CSV.
• Keeps you honest. Guardrails baked in: receipts-only (never fabricate a credential or metric), human-in-the-loop on every send, own the relationship, no single point of failure.

Repo layout

client-sourcing/
├── SKILL.md                       # the skill: the 8-step loop + how to run a sweep
├── references/
│   ├── positioning.md             # YOUR profile, ICP, and proof (a template — fill it in)
│   ├── pitch-playbook.md          # 5 outreach templates + a no-reply follow-up cadence
│   └── channels.md                # where to source, ranked owned→rented, with effort weighting
├── templates/                     # render a sweep into whatever the moment needs
│   ├── email.md                   # Gmail-ready value-first email + follow-up cadence
│   ├── chat-brief.md              # one-screen Slack/Teams brief + weekly roll-up
│   ├── digest.html                # pretty, self-contained HTML digest (open in a browser)
│   ├── salesforce-view.html       # a Lightning Lead-list-view mockup (leave-behind / screenshot)
│   └── crm-salesforce.md          # CRM field mapping + Salesforce import steps
├── scripts/
│   ├── lead_score.py              # 0–100 lead scorer with ICP gates (stdlib only)
│   └── to_salesforce_csv.py       # scored sweep → Salesforce-importable Lead CSV (stdlib only)
├── assets/
│   └── pipeline-template.csv      # the CRM tracking sheet to copy somewhere persistent
├── examples/                      # a full worked run (fictional data)
│   ├── icp-variants.md            # same leads, two positionings, different pipeline
│   ├── leads.scored.csv           # scorer-format input  (lead_score.py --csv)
│   ├── leads.input.csv            # exporter-format input (to_salesforce_csv.py)
│   └── leads.salesforce.csv       # the CRM export it produces
└── docs/                          # rendered previews embedded in this README
    ├── digest-preview.svg
    └── salesforce-preview.svg

Quickstart — the lead scorer

No dependencies; just Python 3.

# score one lead
python3 scripts/lead_score.py --fit 5 --budget 4 --timing 5 --edge 5 --company "Acme Health"

Acme Health — 96.0/100  [A]
  fit 5.0  budget 4.0  timing 5.0  edge 5.0
  → Make the value-first artifact, then a personalized direct touch this week.

# batch-score a CSV, then export it for your CRM
python3 scripts/lead_score.py --csv examples/leads.scored.csv          # scorer columns
python3 scripts/to_salesforce_csv.py examples/leads.input.csv          # → *.salesforce.csv

The hard gates matter: a perfect-timing, deep-pocketed lead that's off-ICP still scores KILL, because chasing the wrong client is the most expensive thing a soloist does.

The "same leads, different practice" demo

examples/icp-variants.md runs three identical leads through two positionings. The ranking reorders — one lead is an A-tier "go" under a privacy practice and an auto-KILL under a marketing one. That's the whole point: your ICP is an input. Tune references/positioning.md and the pipeline follows.

Use it as a Claude skill

• Claude Desktop / Cowork: zip this folder with a .skill extension, then install it via Settings → Capabilities.
• Claude Code: drop the folder where your skills live; the SKILL.md description triggers it on prompts like "who should I pitch this week?" or "draft outreach to ."

Then ask, e.g. "Run a sourcing sweep for this week" and Claude works the loop — source → enrich → qualify → artifact → draft → track.

Make it yours

Edit references/positioning.md — the single source of truth for every pitch and score. Put your one-liner, your ICP, your productized offers, and your receipts (the specific proof that maps to each buyer pain). Ten minutes here is worth more than any other change.

Design notes

• Leading > lagging signals. A job post means the buyer already decided and wrote a spec — late and crowded. Money is upstream.
• Give value before the ask. A small, real, bespoke insight per A-tier target is the reply-rate move. If there's nothing real to give, it isn't A-tier yet.
• Volume is not the channel; quality is. Especially on LinkedIn — automation there risks the ban and the brand. Human-in-the-loop, always.

License

MIT © 2026 Jeff Pinto. See LICENSE.

Privacy

The network features read your LinkedIn Connections export. That file, and everything derived from it, stays local: the email/URL/connected-on columns are dropped at ingest, all generated outputs use gitignored *.local.csv / *.local.md names, and examples/ is 100% fictional. Never commit real connection or prospect data — treat a real name in a commit as an incident.