You are a senior automation engineer. Build the following optimized workflow. # Screen and route inbound applicants Spion recorded this as a manual browser process (6 steps, ~7 min per applicant, dozens per week) and redesigned it as a consolidated automation (4 steps). Build the OPTIMIZED workflow below, not a replay of the manual clicks. Overview: On a new application, parse the resume, score it against the role's must-haves, route qualified candidates to the hiring manager, and send every applicant a prompt acknowledgement. Tools: Greenhouse (or Lever), Claude, Slack Trigger: ATS event (When a new application is submitted for an open role). Efficiency target: 6 manual steps down to 4 (33% fewer), saving ~7 min per applicant, and removing the days-long acknowledgement lag. ## Optimized workflow steps 1. [Greenhouse] Read application What: Read the candidate, resume text, and the job's scorecard / must-have criteria. 2. [Claude] Screen and score (added by Spion: net-new value) What: Extract years of experience, key skills, and location; score against the role's must-haves; write a two-line rationale and a recommend / review / decline call. Why: turns an unread resume pile into a consistent, defensible shortlist — and never on the basis of protected characteristics. 3. [Greenhouse] Tag and route What: Apply the score and stage; advance "recommend" candidates to the hiring manager's review and attach the rationale. 4. [Slack] Notify and acknowledge What: DM the hiring manager a digest of new "recommend" candidates; send every applicant a same-day acknowledgement email. ## Original manual process (reference only, do not replicate) 1. Open each new application 2. Skim the resume 3. Judge it against the job criteria 4. Move it to a stage 5. Ping the hiring manager 6. Send (or forget) an acknowledgement ## What I need from you 1. Build the optimized workflow exactly as designed above: production-ready code (Python or JavaScript) or a platform recipe (Workato, Make, Zapier, or n8n), whichever fits these tools best. 2. Preserve the consolidation. Do not expand it back into the manual click-by-click process. 3. Recommend the right trigger and cadence (schedule, webhook, or event). If it recurs, state the exact schedule. 4. Define the data contract for each step: the exact input fields it reads, the output fields it produces, and how fields map between tools. 5. Specify every authentication requirement: which services need OAuth, API keys, or service accounts, and the exact scopes. 6. Score only on job-relevant criteria. Never use or infer protected characteristics (age, gender, race, religion, national origin, disability), and keep a human as the final decision-maker on every reject. 7. Add robust error handling (missing resume text, auth failures, empty results, rate limits with retries) and make the workflow idempotent so it is safe to re-run. 8. Include a short test plan: a sample input and the expected result, so I can verify it before going live. 9. Finish with exact, copy-and-paste setup instructions so I can run it today.