joseym/aarg

AARG: The Adversarial Agentic Resume Generator

AARG tailors your résumé to a specific job posting and then argues with itself about the result. A skeptical reviewer agent reads each draft the way a hiring manager looking for reasons to pass would, files specific objections, and a tailoring agent revises against them under tight bounds, keeping the best draft it finds rather than the last. The output is two PDFs built with Typst: a plain, parser-safe one to upload, and a designed one to hand a person.

It runs on your machine, against your own career data, and it will not invent experience you don't have. That last part is the whole point, and it's enforced in three places rather than promised in one.

Status: working end to end. Ingest, tailoring, the adversarial loop, both résumé variants, history/diff, and an interactive shell all work today. Anthropic is the supported model provider; a fully-local one is on the roadmap, not in the box yet.

The loop, briefly

1. ingest turns an existing résumé into a structured dataset: roles, bullets, and skills, each tied to evidence.
2. tailor <job> parses the posting, runs a gap analysis against your dataset, and writes a first draft that mirrors the posting's language without overstating what you've actually done.
3. The adversarial reviewer scores the draft on content and on deterministic keyword coverage, then files objections: no metric, vague verb, unsupported claim, and so on.
4. The tailoring agent revises against those objections and re-scores. A revision that doesn't improve the score is discarded and the loop stops; the build keeps the best draft it ever saw.
5. The winner renders to an ATS PDF and a human PDF (same facts, different presentation), and every iteration is written to disk so you can inspect or diff it later.

When an objection can't be satisfied without lying ("this bullet states an outcome with no number"), t