Drop-in patch over v0.1.0. No breaking changes.

What's new

• 22-second demo video at the top of the README — Remotion-rendered MP4 + GIF showing the three non-compete scenarios cycling, with the operand trace revealed on the failing requirement. Stills extracted for inline use.
• Humanized landing page — stripped AI tells (em-dash overuse, rule-of-three, 'testament / landscape / shoulders of giants' vocab) and put back a human voice.
• GitHub icon in the nav — real SVG, not just a word.
• Packaged Claude Skill — nomos-reason.zip attached to this release. Drop into ~/.claude/skills/ and Claude starts writing .nomos programs.
• 42 tests covering the new grammar: negation, !=, duration literals, user-defined predicates, operand surfacing.
• MODEL_ALIASES refreshed against live OpenRouter (Sonnet 4.6, Opus 4.1–4.7, Haiku 4.5, GPT-5 variants, Gemini 3 previews).

Install the Claude Skill

curl -L -o /tmp/nomos-reason.zip \
 https://github.com/sboghossian/nomos/releases/download/v0.1.1/nomos-reason.zip
mkdir -p ~/.claude/skills
unzip -o /tmp/nomos-reason.zip -d ~/.claude/skills/

Restart Claude Code. Ask it to write a Nomos program.

Full changelog

See CHANGELOG.md and /changelog.

Nomos is a programming language for legal reasoning. v0.1.0 is the first public release. Side project, built in public, Apache-2.0.

• 🌐 Site: https://nomos.dashable.dev
• 🕹 Playground: https://nomos.dashable.dev/play
• 📊 CUAD benchmark: https://nomos.dashable.dev/research/benchmarks

What works end-to-end

• Lexer + parser + typed AST, with 26 Vitest specs pinning behavior
• Name resolver with diagnostics
• Expression evaluator + rule engine with temporal validity
• Defeasibility solver: priority → specificity (lex specialis) → recency (lex posterior) → declaration order, with human-readable tiebreaker explanation
• Typed LLM bridges: extract<T> using llm(...) verified_by human if confidence < N calls OpenRouter, returns typed values, routes low-confidence results to a human queue
• Typed AuthorityRef AST: authorities classified as article / case / section / decree / generic with canonical strings
• Eyecite (Free Law Project) integration — nomos resolve resolves US citations to structured records (Brown v. Board → 347 U.S. 483 / 1954 / reporter U.S.)
• CLI with colored sectioned output: nomos run/parse/check/resolve
• VS Code extension with TextMate grammar + LSP (diagnostics, hover, completion)
• Web playground running the full pipeline client-side + proxy API for LLM calls (rate-limited, daemonized via launchd)
• Claude Skill ready for lawvable/awesome-legal-skills submission

CUAD benchmark (Claude Sonnet 4.5, 20 samples)

Honest — reproducible via bench/cuad/harness.mjs.

Credits

Fifty years of rules-as-code research: Catala (Inria), OpenFisca, Blawx, Logical English (Kowalski), LegalRuleML, and the 1981 British Nationality Act encoding. Nomos sits on top — see /research/prior-art.

Not

Not a law firm. Not legal advice. Not production software. Experimental. Breaking changes expected.