requisition-intelligence

Requisition Intelligence

Paste a job description, get a full sourcing toolkit back. Paste a candidate profile, get a fit evaluation. Ask for interview prep, get questions a recruiter can actually assess the answers to.

Built by a senior technical recruiter, iterated across five versions in real searches, and shared open source for the TA + AI builder community. Designed for technical recruiters hiring software, infrastructure, systems, and cloud engineers in fast-paced environments — the gap this closes is the one between "JD lands in your inbox" and "you're actually sourcing".

Two ways to run it

Both versions are the same methodology.

Setting up the Gem

1. Create a new Gem in Google Gemini and paste everything below the first horizontal line of GEM-PROMPT.md into the instructions field.
2. Customise the "Context & Identity" section first — this is mandatory. The prompt cannot produce useful output without your company name, culture, hiring locations, compensation strategy, ATS, and the types of roles you hire. Every placeholder in square brackets needs replacing with your own details.
3. Phase 1 fires when you paste a job description — full analysis and sourcing toolkit.
4. Phase 2 activates when you paste a candidate profile after a JD has been analysed — a fit evaluation. Always apply your own judgement: it's a second opinion, not a decision-maker.
5. Phase 3 triggers when you ask for interview questions or want to refine your ATS screening gate.

The Claude skill collects the same company context conversationally on first use — install it and paste a JD.

How it works

flowchart TD
    A["Paste a job description"] --> P1["Phase 1 — Requisition toolkit"]
    P1 --> B1["Breakdown, hidden requirements, red flags"]
    P1 --> B2["Hiring manager clarification questions"]
    P1 --> B3["Realistic 7-8/10 candidate profile"]
    P1 --> B4["Boolean strings + alternative sourcing channels"]
    P1 --> B5["Outreach, follow-up + LinkedIn post"]
    P1 --> B6["ATS screening gate (top ~15%)"]
    C["Paste a candidate profile"] --> P2["Phase 2 — Fit evaluation"]
    P2 --> D1["Fit, gaps, screening angle, verdict"]
    E["Ask for interview prep"] --> P3["Phase 3 — Screening prep"]
    P3 --> F1["Recruiter-calibrated phone screen questions"]
    P3 --> F2["ATS gate refinement"]

Phase 1 — from JD to sourcing execution

One paste produces ten outputs: a plain-English requisition breakdown, hidden requirements and red flags, ready-to-send hiring manager clarification questions, a realistic strong-but-not-unicorn candidate persona, a competitive positioning note, LinkedIn Boolean strings (primary + broadened variant), role-specific free sourcing channels beyond LinkedIn, a character-counted InMail and follow-up, an inbound-generating LinkedIn post, and a set of ATS screening questions engineered with compound pass rates to gate to roughly the top 15% of applicants.

Phase 2 — sourcing soundboard

Paste a profile after the JD has been analysed and get a strict evaluation against that requisition: fit rating, alignment breakdown against the must-haves, gaps and risks stated plainly, and a clear pursue / caution / pass verdict. Always apply your own judgement — this is a second opinion, not a decision-maker.

Phase 3 — interview prep for non-engineers

Every question is designed so a technically literate recruiter — not an engineer — can meaningfully evaluate the answer: decisions and trade-offs, scale and scope, recency and depth, with strong-answer and weak-answer signals described for each. Includes recalibration of the ATS gate once real applicant data comes in.

Design principles

• Free sourcing channels only — every recommendation must pass the test: can you open it in a browser and find sourceable candidates within 15 minutes?
• Hard character limits on outreach (650), follow-ups (250), and posts (1,300) — verified before output, never exceeded.
• Realistic candidate profiles — sourceable 7-8/10 personas, not unicorn checklists.
• Strict gates that are hard to fake — screening questions force specificity (scale, duration, recency) over self-assessment, including one honesty-check question per set.
• Calibrated to the recruiter, not the engineer — no output requires engineering knowledge to use.

Contributing

Fork it, improve it, share what you learn — see CONTRIBUTING.md. The CHANGELOG shows how it has evolved across five versions of real-world use.

Licence

MIT — see LICENSE. In plain English: you can use, copy, modify, and share this freely, including commercially, as long as the licence notice stays with it. No warranty.