1. Shares the same "CogniAgent Recruiting Data" workbook as the Candidate Screening Agent, creating it if it does not exist yet. Install either one first — they converge on the same six tabs. 2. Takes one candidate marked for qualification, claims the row so no two runs can pick up the same person, and reads the competencies for that job from the Competencies tab. 3. Emails the candidate a structured set of questions covering each competency, asking for concrete examples rather than opinions. 4. Scores each answer against your guide. Every score must quote the candidate's own words as evidence — a score whose evidence is not actually in what they wrote is rejected and flagged for a human to read instead. 5. Sends the recruiter a summary: the score per competency, the evidence behind it, the gaps, and a recommendation to advance, hold or decline. 6. The recruiter accepts or overrides. Their decision is what gets written and what the candidate is told — the agent's recommendation never reaches the candidate on its own. Runs on its own, or after the Candidate Screening Agent — that one hands over by marking the row, so no extra wiring is needed.