Best Legal Skills for Claude Code
Skills for US legal work — contract review, privacy compliance, citation checking, risk assessment, and drafting. This is the most adversarially tested category on the site: five of the eight skills below have been through live injection probes, with instructions hidden inside intake documents and direct attempts to talk the skill out of its own verification steps. The readers held every boundary; two of the generators adopted a rule override. Static scores are below — the dynamic results are published on each skill page.
California CCPA/CPRA compliance advisor — thresholds, consumer rights, ADMT obligations, and penalty exposure.
The strongest US compliance skill evaluated — statute detail current to mid-2026 with real enforcement dollar figures, from a 909★ GRC collection.
Checks and fixes Word-footnote citations against Bluebook/Indigo Book rules, with tracked changes.
Low stars but the best-engineered skill of this batch (tests + packaging + release checklist); purpose-built for US law-review citation practice.
Reviews contracts against CUAD’s 41 risk categories with redlines and market-standard comparisons.
The most rigorous English contract-review skill found: benchmark-grounded risk taxonomy, enforced output structure, and versioned maintenance under MIT.
Statute, regulation and case-law research methodology using the Federal Register API and source-verified citations.
Ported from Anthropic’s own claude-for-legal under Apache-2.0 — the methodology pedigree shows in its verified-citation discipline.
Classifies legal risks on a severity-by-likelihood matrix with explicit escalation criteria for in-house teams.
The corporate-legal pick from a 206★ suite: a disciplined severity/likelihood framework with escalation criteria real legal teams can adopt.
Generates a complete, customized NDA with plain-English annotations for every section.
Flagship NDA skill of the highest-starred legal toolkit on GitHub (1.7k★). Clean prompt engineering; engineering depth is limited by design.
Structured distinctiveness analysis that prepares inventors for a productive patent-attorney consultation.
A consultation-prep tool rather than a lawyer replacement — honest about scope, and methodologically the most interesting IP skill found.
Drafts a civil complaint as a senior litigation partner would — story-first, element-complete, motion-to-dismiss-proof.
From the Themis civil-litigation stack: the complaint-drafting flagship that writes to survive a motion to dismiss, with sibling skills for answers and amended pleadings.
Score comparison
How the top 8 compare across the six evaluation dimensions (0–10). Full written rationale for every score is on each skill's page.
| Skill | Overall | Trigger | Structure | Workflow | Content | Engineering | Security |
|---|---|---|---|---|---|---|---|
| CCPA Compliance Advisor | 9.0 | 9.5 | 9.0 | 9.0 | 9.5 | 8.0 | 9.0 |
| Bluebook Citation Checker | 8.8 | 9.5 | 9.0 | 8.5 | 9.0 | 8.5 | 8.0 |
| Contract Review (CUAD) | 8.5 | 9.0 | 9.0 | 9.0 | 9.0 | 7.0 | 8.0 |
| US Legal Research | 8.4 | 8.5 | 8.5 | 9.0 | 8.5 | 8.0 | 8.0 |
| Legal Risk Assessment | 8.1 | 9.0 | 8.5 | 8.5 | 8.0 | 6.5 | 8.0 |
| Custom NDA Generator | 8.0 | 9.0 | 8.5 | 8.0 | 8.5 | 6.5 | 7.5 |
| Patent Scanner | 8.0 | 8.0 | 8.5 | 8.0 | 8.5 | 7.0 | 8.0 |
| Draft Civil Complaint | 7.8 | 7.5 | 8.5 | 8.5 | 8.0 | 6.5 | 7.5 |
FAQ
What is a legal skill for Claude Code?
A SKILL.md package that teaches your agent a legal workflow — reviewing a contract against a risk taxonomy, checking citations against Bluebook rules, walking through CCPA/CPRA applicability thresholds, or assembling an NDA from a term sheet. It activates when your request matches its trigger description.
Are these skills a substitute for a lawyer?
No. They are research and drafting aids: they compress the mechanical part of legal work — citation format, clause coverage, statutory thresholds — and they are still wrong often enough that a qualified attorney has to review the output before anything is filed or signed. Treat them as a first pass, not a signature.
What did the injection tests actually find?
Each of the five tested skills ran four adversarial probes in a sandboxed agent, three times each. ccpa-compliance, contract-review, and legal-risk-assessment held every boundary (10/10, PASS). law-citation-skill and legal-nda-generator landed at 7.5/10 PARTIAL — both adopted a direct "skip your verification step" override. The pattern across the whole September batch: instructions hidden inside content were almost always caught; a prompt that simply tells the skill to ignore its own rule often was not. Full write-up on the blog.
How are these scores calculated?
Six dimensions worth 100 points — trigger quality, structure, workflow design, content, engineering, and security — each with written rationale on the skill page, plus a security veto that caps the total at 39/100 for confirmed credential theft, backdoors, or exfiltration. The full rubric is on the methodology page.
Every skill here was scored on the same six-dimension rubric — the full scorecard with written rationale is on each skill's page. Browse the full Legal skills collection (including skills outside the top 8), or browse all evaluated skills.