RECRUITER.md
Sourcing and outreach that stays inside your InMail budget and does not infer things it should not. Written for the assistant doing the sourcing: requirements split into what a profile can evidence and what it cannot, the credit budget the work has to fit, and explicit fairness constraints — because a model left to its own devices will happily rank on graduation year.
# Recruiting — <role>
## The role
<Senior Backend Engineer (Go), hybrid London 2 days, £95–115k, reporting to the VP Engineering.>
## Requirements
**Evidenceable from a LinkedIn profile:**
| Requirement | Must / strong / nice | What counts as evidence |
|---|---|---|
| <Go in production> | <must> | <named in a role description or the skills section> |
| <Payments domain> | <strong> | <worked at a payments company or named it in a role> |
| <Team of 3+ led> | <nice> | <stated explicitly; do not infer from a title> |
**Not evidenceable — settle on the call:**
<- Communication
- Whether they want this kind of team
- Why they left the last role>
Never score the second list from a profile.
## Fairness constraints
The assistant must not infer or use: gender, age, ethnicity, nationality, religion, health, disability or family status — nor proxies for them, including graduation year, career gaps, name or photo. If a career gap is relevant to a requirement, note it neutrally as a fact for a human to ask about; do not score it.
Jurisdictions differ on automated screening: New York City requires a bias audit for automated employment decision tools, and the EU AI Act treats recruitment screening as high-risk. This file assumes the assistant produces evidence for a human decision and never makes the decision.
## Budget
- Plan: <Recruiter Lite>
- InMail credits per month: <30>
- Search result cap: <1,000>
See LINKEDIN-LIMITS.md. A shortlist longer than the credit budget is not a shortlist.
## Sourcing map
<See the talent map output. Archetypes, the titles the role is called at each, and the search per archetype.>
## Outreach rules
- Name the company in the first message. Withholding it halves the reply rate.
- Give the salary range if permitted. If not, say explicitly that you cannot yet and why.
- The ask is a question, not a call.
- One follow-up, then stop.
- Never use the candidate's current employer's problems as leverage.
## Disclosure
<What may be said about the team, the funding, the reason the role is open, and the reason the last person left. Be specific about what is confidential — an assistant told nothing will either say nothing or guess.>- Claude Project knowledge
- ChatGPT Project files
- Agent folders
- The role, the requirements split, and the disclosure rules
Everything in <angle brackets> is example content, not instructions. Replace it. The file ships filled in rather than empty on purpose: an empty template gets abandoned, a filled one gets edited.
The reference behind this file
Where the 30-credit and 1,000-result figures come from, and what Corporate changes.
Written to a fixed structure and run through Claude and ChatGPT to confirm each one returns the output format it declares. That is not the same as a measured reply-rate benchmark, and nothing here claims one — if you run these at volume and have numbers, we want them.
Packs that already contain this file
A pack is the folder with this file, the ones it references and the prompts that use them — so you are not assembling fifteen downloads by hand.
One file is a start. The folder is the point.
Files reference each other — CLAUDE.md declares the reading order, the prompts expect the files by name. The packs ship them together, already wired up.