Cv Screener

CV Screener & Ranker

What This Skill Does

Takes a batch of CVs + a role definition and returns a ranked shortlist — each candidate scored, with a one-line reason and any red flags. The point is to get the user to a confident "interview these N" decision fast, while flagging the close calls a human should look at.

This is a first-pass filter, not a final decision. It surfaces signal; the human decides.

---

Process

1. Get the role bar. Pull must-haves, nice-to-haves, and red flags from the JD or rubric. If

none provided, ask for the 3–5 things that genuinely matter for this role.

1. Read each CV against the bar. Score must-haves (pass/partial/miss), note nice-to-haves, flag

red flags.

1. Look past keywords for real signal: evidence of slope (shipped things, fast growth, self-

driven projects), trajectory, and whether their experience maps to this problem vs. a generic match.

1. Rank into tiers: Strong yes / Worth a look / Likely no — with reasoning.
2. Flag close calls explicitly so the human reviews them rather than trusting the cut.

---

How to Screen Well

• Slope over pedigree. A candidate who shipped a side project and grew fast often beats a glossier

résumé. Reward evidence of self-driven improvement.

• Map to the actual problem. "5 years in X" matters less than whether their experience fits this

role's real challenge. Note when a strong-looking CV is a weak fit and vice versa.

• Be honest about gaps. Don't pad a thin CV to fill a shortlist. A short honest shortlist beats a

padded one.

• Surface, don't bury, red flags. Job-hopping with no narrative, unexplained gaps, claims that

don't add up — name them. The human decides if they matter.

• Never auto-reject on proxies that correlate with protected characteristics (school prestige,

name, gaps that may be caregiving). Score on demonstrated capability.

---

Output Format

# CV Screen: [Role] — [N] candidates

## 🟢 Strong Yes
| Candidate | Fit | Why | Watch |
|---|---|---|---|
| [name] | [score] | [one line] | [any flag] |

## 🟡 Worth a Look
[same table — the close calls; note what to verify in a screen]

## 🔴 Likely No
| Candidate | Why not |
|---|---|

## Notes
[Anything the human should know — e.g., "3 candidates were borderline on X; I'd screen them rather
than cut." Or "the strongest CV on paper is a weak fit for this specific problem."]

---

Screener Failure Checklist

A screen FAILS if it:

• ❌ Ranks on keyword matching instead of real fit to the role's actual problem
• ❌ Rewards pedigree over demonstrated slope
• ❌ Pads the shortlist to hit a number instead of being honest about a thin field
• ❌ Hides red flags or close calls instead of surfacing them for human review
• ❌ Auto-rejects on proxies for protected characteristics
• ❌ Gives scores with no reasoning the human can sanity-check

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: getclera
• Source: getclera/claude-skills
• License: MIT

Install and usage instructions live in the source repository linked above.