What’s inside

One SKILL.md (the playbook) + one starter template. A preview:

Resume Screener

When to use

• You have 50+ applications for one role and no time to read them all
• You need a consistent, written scoring rubric applied to every candidate equally
• Hiring runs on a schedule and you want a digest of new applicants each cycle
• You want public evidence (GitHub, portfolios, published work) checked against resume claims
• Your team needs a defensible shortlist with reasons attached to each name

Workflow

Turn the job description into must-have and nice-to-have criteria with weights — e.g. "5+ years backend: required; fintech experience: nice-to-have." This is the most important step: a sloppy rubric produces a sloppy shortlist. Include how borderline cases should be flagged, not just how winners score.

1. Write the rubric (input: job description → output: scoring rubric)

Gather all resumes into one place: a folder, an inbox, or a scheduled pull from your ATS. One candidate, one file or record — name them consistently so nothing gets lost or double-counted.

1. Collect the pile (input: applications → output: resume set)

Parse each resume for name, years of experience, and key skills, then score it against your rubric. Where public evidence exists (GitHub, portfolio, writing), check the claims against it — a resume line with evidence behind it scores higher than one without. Every score carries a one-line reason.

1. Score every candidate (input: rubric + resume set → output: scored candidates)

Before finalizing, pressure-test the top of the list: have a second pass argue against the highest scores. Is the evidence real, or is the resume just well-written? Force a real score distribution so "everyone looks good" can't happen. This step is what separates a ranking from a rubber stamp.

1. Challenge the top (input: scored candidates → output: defended ranking)

Deliver a ranked shortlist where every candidate has a score and a reason. Borderline cases get explicit flags — "strong portfolio, but only 3 years experience vs. required 5" — so you can decide whether the rule bends. Nothing ambiguous gets silently dropped or silently promoted.

1. Flag the uncertain (input: ranked list → output: shortlist + flags)

You review the shortlist and make the interview call. The AI never contacts candidates directly — interview invites and rejection notes get your approval first. Final hiring decisions stay with humans, always.

1. You decide (input: shortlist → output: interview list)

Output spec

• One score per candidate per rubric criterion, plus a total and a rank
• One-line reason attached to every score
• Borderline cases flagged with the exact reason they're uncertain
• Ranked shortlist table: name, total score, top strengths, biggest risk
• Digest mode: the same output d

Download the full skill pack (ZIP)