i.
Extract the rubric
Claude / Codex read the JD and pull out a structured rubric: must-haves, nice-to-haves, implicit signals (years, scope, domain). Each requirement gets a weight.
ii.
Score each axis
For each rubric item, the agent searches your resume for evidence and scores the match: full / partial / gap. Returns structured JSON, not vibes.
iii.
Suggest the edits
For weak axes, the agent recommends specific resume rewrites — "lead with this metric," "reframe this title," "this skill is in your work but not on the page."