About Domain Experts
Open source library of job role definitions — the actual mental models, decision thresholds, and failure modes of real practitioners, structured so any AI agent can load one and reason like that expert. Ask your agent to "review this contract" and it answers with a senior contracts attorney's clause playbook and fallback ladders, not a generalist's summary of the internet.
npx domain-experts match "review this vendor contract like a lawyer"
npx domain-experts add lawyer-contracts # installs into ./.claude/skills/"Act as a contracts lawyer" → generic checklist, no thresholds, no fallback order, different answer every run.
Loads lawyer-contracts → indemnity cap red-flag at <1x fees, liability fallback ladder (mutual cap → carve-outs → walk), same reasoning every time.
Why not just prompt "act as a CFO"?
You'll get the average of every job description on the internet, regenerated from scratch each session, verified by no one. These roles pass a non-derivability test — numeric red-flag thresholds, worked examples with reconciling numbers, fallback positions in preference order — built through an adversarial-critique pipeline and CI-linted on every PR.
How it's structured
Each role is a compact reasoning core (SKILL.md) plus on-demand depth (references/: artifacts, red flags, vocabulary). Works in Claude Code, Codex, Cursor, and 30+ agent tools that read the SKILL.md format.
How we verify
Not just "written by experts" — evals/run_evals.py measures counterfactual wins (skill vs. generalist baseline) and parity against real practitioners' public answers, reproducible by anyone.
Full pipeline, evidence, and per-tool install instructions: README on GitHub.