AnythingButLaw (法外功夫)
Nine non-legal business analysis frameworks for lawyers — decision, negotiation, due diligence, valuation, and statistical evidence — packaged as one AI skill.
Pure markdown knowledge skill with no network calls or sensitive data handling; however frontmatter declares allowed-tools: Read, Write, Edit, Bash — Write/Edit/Bash are excessive for a reference-only skill and unjustified, violating least privilege; publisher unverified. Deducted for overly broad tool grants and absence of confirmation/rollback design.
Routing logic is clear, INSTALL.md provides a verification prompt and expected behavior, structure is self-consistent with no dependencies; but references/ and examples/ files are absent from this evidence set, key paths unverifiable statically, no tests or abnormal-input behavior described. Deducted for missing core files and no failure-feedback design.
Audience (Chinese lawyers) and scenarios (settlement, due diligence, antitrust, statistical evidence) are explicit with detailed trigger conditions in the description and bilingual output; but non-fit boundaries are undeclared — 'acknowledge limitations' is the only hedge for cases needing professional judgment. Deducted for incomplete non-fit and false-trigger safeguards.
Well-layered docs (SKILL.md routing + on-demand references + examples), MIT license, version and changelog present, bilingual READMEs; but no maintenance ownership or update path, and minor inconsistency between README file listing (README_ZH.md) and actual files. Deducted for missing governance and structural inconsistency.
Clear design goal; worked examples and adversarial checklists offer framework value beyond generic Q&A; README claims 100% pass rate vs 73.3% baseline but evaluation method, sample and scoring are unauditable, at +66% tokens and +47% time cost. Deducted for uncorroborated benefit claims and meaningful cost increase.
Evaluation results are author self-report only, with no scripts, scoring rubric, or raw records; cited case law and the inspiration link cannot be verified within this evidence set. Deducted for unverifiable core effectiveness claims.
- Frontmatter grants Write/Edit/Bash though content is reference-only; recommend narrowing to Read with justification.
- Actual content of references/ and examples/ was not provided in this evidence set; verify formulas and case citations before relying on them.
- The README's evaluation numbers (100% pass rate) lack methodology and should not drive adoption.
- Output is an analytical framework, not legal advice; quantitative conclusions in real matters require licensed-lawyer review.
- Publisher identity is unverified with no stated maintenance or update commitment.
What does this skill do, and when should you use it?
Law school teaches law, but client problems rarely stop at legal questions. This skill distills nine analytical frameworks — decision analysis, game theory, contract design, accounting, finance, microeconomics, law & economics, statistics, and multivariate regression — into deep, lawyer-oriented reference material. Every domain ships with LaTeX formula derivations, fully worked numerical examples, adversarial checklists, and landmark cases. The author assembled it from his JD experience at Notre Dame and open-sourced it as a substitute for a curriculum he estimates at roughly 2 million RMB.
The skill loads 9 reference files and 10 complete case studies, each containing formulas, calculations, sensitivity analysis, and recommendations. Claude routes user questions via built-in routing logic: 'should we do X' decisions go to decision analysis, settlement talks add game theory, due diligence red flags go to accounting, damages and valuation go to finance, and regression challenges go to multivariate statistics. Output follows a defined style: LaTeX notation, bilingual terminology, comparison tables, ranges with key thresholds instead of point estimates, adversarial-checklist questioning, an overall framework first for cross-domain problems, and explicit acknowledgment when data is insufficient.
- A litigator weighing a settlement offer: decision trees, expected value, Bayesian updating, and VOI applied to a 1.1M offer vs. trial with punitive-damages tail risk
- A commercial lawyer drafting a service contract: principal-agent incentives, hold-up, and relational-contract risk of fixed-fee vs. cost-plus structures
- An M&A due diligence team that can't read the CFO's numbers: DuPont, Altman Z, Beneish M, and revenue-recognition red flags
- An antitrust lawyer running a merger review: SSNIP, HHI, and elasticity analysis for market definition
- An employment lawyer challenging opposing regression evidence: omitted variables, reverse causality, IV and DiD tools
- A tort lawyer handling product liability: the Hand formula as the economic standard for negligence
What are this skill's strengths and limitations?
- Complete nine-domain system, each with formula derivations, worked examples, adversarial checklists, landmark cases, and pitfall quick-reference tables
- 10 fully worked numerical case studies, single-domain and multi-domain, grounded in realistic legal scenarios
- Bilingual terminology throughout, lowering the barrier for Chinese-speaking practitioners
- Author publishes evaluation results: pass rate improved from 73.3% to 100% with the skill (at the cost of +66% tokens and +47% latency)
- MIT licensed, with an explicitly specified output style (sensitivity ranges, admitted limitations, adversarial questioning) rather than generic lecturing
- The evaluation's test set, sample size, and methodology are not disclosed in the source material — the 100% pass rate cannot be independently verified
- Content is framed for legal practice and is not a substitute for formal opinions from accountants, economists, or statisticians
- Built against the Claude Code skill standard; behavior on other Agent Skills clients is unverified
- Much of the content (law & economics, U.S. precedents like Daubert) is anchored in U.S. law; applying it directly to Chinese practice requires reader judgment
- Token consumption and response time are notably higher than baseline (+66% / +47%), which matters for heavy usage
How do you install this skill?
git clone https://github.com/TracyWang95/AnythingButLaw.git ~/.claude/skills/AnythingButLaw
The entry point is SKILL.md (v2.0.0); the bulk of content lives in references/ and examples/. The repo also contains INSTALL.md and README_EN.md, which the author points to for details.
How do you use this skill?
Triggers automatically once installed — no manual invocation needed. Example prompts: "Client received a 1.1M settlement demand. 50% liability probability, 1M expected damages, but 10% chance of a 25M punitive award. Should we settle?" or "Opposing counsel filed a regression controlling for education and tenure claiming the gender coefficient is insignificant. How do I challenge it?" Complex cross-domain questions (e.g., an M&A deal involving valuation + contracts + game theory) read multiple reference files.
How does this skill compare with similar options?
The author explicitly cites @titanwings' colleague.skill as the inspiration; the other reference point in the README is the U.S. law school course on non-legal lawyer skills that this skill replaces.