Screen 300 applicants to a real shortlist

Founder/hiring manager at a 5-50 person startup with no recruiter · Assemble in ~20 min; first scored shortlist from a 100-application export in 30-60 min depending on batch size.

HiringEdit flowon X◆ new (unverified) 83

step 1ready

Define the scoring rubric

hiring10 alts

step 2connector

Export the applicant pile from your ATS

⚡ bring your own tool

step 3partial

De-duplicate and filter ineligible applications

job-pipeline1 alts

step 4partial

Flag AI-generated and fabricated resumes

avoid-ai-writing

step 5ready

Score each resume against the rubric

resume-evaluator

step 6ready

Adversarially verify top-candidate red flags

adversarial-verify

step 7partial

Draft advance and rejection emails

excel-fire-recruiting9 alts

Did this recipe work for you?

What you get

▸Ranked candidate table with numeric scores, rubric traces, and per-candidate scorecards
▸AI-fabrication flag report: candidates marked high/medium/low fabrication risk with evidence
▸Adversarially verified red-flag list (only surviving, well-evidenced concerns)
▸Ready-to-send advance invitation and rejection email drafts, one per candidate, personalized

tools: Your existing ATS (free tier on Ashby/Workable sufficient for export) — no paid add-ons required

tokens: ~$4-10 per 100-application batch (scoring + AI-flag pass + email drafts)

≈ One-time or per-hire cost, not recurring — $5-15 per hiring round at normal startup cadence

What stays yours

· Review all AI-flagged resumes before disqualifying — the flag is a signal, not a verdict; legitimate candidates sometimes use AI writing tools
· Read the top-10 scorecards and make the final shortlist call — the agent scores, the founder decides
· Approve and send every email draft — no auto-send; employment law in most jurisdictions holds the employer responsible for rejection communications

The problem

When my job post gets 300+ applications (half AI-generated), I want every resume scored against my actual must-haves with evidence and the fakes flagged, so that I only spend founder time on the top 10 real candidates.

Cybernetic layers

Memorybyo

Candidate state (who was contacted, response received, moved to next stage) lives in your ATS or a tracking spreadsheet. No library memory artifact covers this yet. Export a running CSV and pass it back in if you run a second batch.

Background-schedulingnot-applicable

Inbound screening is a batch job triggered by the founder, not a continuous loop. Run it once per posting close date. No cron needed; scheduling it via Claude Code background task adds no value here.

Harnessready

Claude Code is the runtime. Install each skill into ~/.claude/skills/ and invoke by name in sequence. The subagent (resume-evaluator, id 7516) and the workflow (job-pipeline, id 5758) are invoked as Claude Code agents — no extra harness required.

Click any step to see what fills it — and the ranked alternatives you can swap in.