AI Legal Assistant (legal/SKILL.md Orchestrator)
One /legal command in Claude Code for contract review, risk scoring, NDA and terms generation, compliance checks, and PDF reports — replacing costly first-pass legal review.
The SKILL.md itself requests low privileges (file reads, WebFetch, local Markdown writes) with transparent data flow, and mandates a legal disclaimer on every output plus attorney referral — positive signals. Deducted for: no stated handling boundaries for sensitive contract text, no rollback mechanism, no data-flow disclosure for WebFetch of external URLs, and unverifiable details around the curl|bash install path.
Routing table is clear and missing-input handling asks the user explicitly; the main path (read → analyze → write Markdown) is self-consistent. Deducted for: no tests or sample inputs/outputs, PDF dependency (reportlab) failure feedback not covered in this file, and no guidance for abnormal inputs (corrupt, oversized, non-English contracts).
Audience (freelancers, small businesses) and scenarios are clear; command semantics and trigger conditions (/legal prefix) are precise with low false-trigger risk. Deducted for: undeclared capability boundaries (jurisdictions, languages), fully English output, and total dependence on overseas services (Claude Code, WebFetch) with no Chinese support, raising mainland-China usability concerns.
Structure is clean (command menu, routing table, naming conventions, output naming scheme) and readable. Deducted for: unknown license metadata, no versioning, changelog, or maintenance ownership, marketing metrics in README without sources, and mixing of Skool community promotion that dilutes documentation integrity.
The promised core flows (contract analysis, document generation, Markdown output) are plausible at the instruction level with directly usable output formats. Deducted for: key value claims (Contract Safety Score, severity scoring, GDPR/CCPA coverage) rest entirely on model quality, with no representative output samples; static review cannot confirm real usability.
Only author claims and marketing statements ('review in 60 seconds', dollar-savings metrics) with no sources; no test suite, CI evidence, or third-party execution material; facts and inference are not separated. The only auditable material is the repository source itself.
- The disclaimer is not legal protection: AI-generated contract analysis and documents must be reviewed by a licensed attorney; risk flags here are not legal opinions.
- License metadata is unknown; confirm terms before any commercial use.
- No versioning or maintenance commitment; users bear update and abandonment risk.
- Treat quantified outputs (Contract Safety Score, severity scoring) as heuristic references, not a validated assessment system.
- Core function depends entirely on overseas services (Claude Code, WebFetch); mainland-China reachability is unverified and there is no Chinese-language output support.
- Publisher is unverified in the FollowSkills registry; identity is unknown (not itself a deduction, but disclosed).
What does this skill do, and when should you use it?
This is the main orchestrator skill of the zubair-trabzada/ai-legal-claude repository, located at legal/SKILL.md. It performs no analysis itself; instead it routes 13 /legal slash commands to concrete skills in the repo's skills/ directory (legal-review, legal-nda, legal-compliance, etc.). The flagship /legal review command launches 5 parallel agents (clause analysis, risk assessment, compliance, obligations mapping, recommendations) that aggregate into a 0–100 Contract Safety Score with clause-by-clause analysis. Every output carries a mandatory legal disclaimer stating the tool is not a lawyer and provides no legal advice. A one-line curl installer deploys all 14 skills, 5 agents, and the PDF generation scripts.
When the user types /legal, it presents a 14-command menu and routes each command: /legal review (5-agent parallel full contract review), /legal risks (clause-by-clause severity scoring), /legal compare (side-by-side diff of two contracts), /legal plain (legalese-to-plain-English translation), /legal negotiate (counter-proposals with replacement language), /legal missing (finds absent protections), /legal nda / terms / privacy / agreement (generate NDAs, terms of service, privacy policies, business agreements), /legal freelancer (freelancer-perspective review), /legal compliance (GDPR/CCPA/ADA/PCI-DSS/CAN-SPAM/SOC 2 gap analysis), /legal report-pdf (ReportLab-based PDF report). It accepts contracts as a file path, pasted text, or URL (via Read/WebFetch), and saves generated documents as Markdown in the working directory (e.g., NDA-[party]-[date].md).
- A freelancer reviewing a client contract before signing to spot common contractor trap clauses
- A small business assessing vendor or supplier contracts and estimating financial exposure before committing
- A founder generating a custom NDA for a new partnership, or terms of service and a privacy policy for a website
- An AI automation agency packaging contract review as a paid service with client-ready PDF reports
- A site operator running a compliance gap check against GDPR, CCPA, ADA, PCI-DSS, CAN-SPAM, or SOC 2
What are this skill's strengths and limitations?
- Command coverage spans the full contract lifecycle: review, compare, negotiate, generate, comply, report
- Flagship /legal review cross-checks with 5 parallel agents and outputs a structured 0–100 safety score with per-clause fixes
- Requires explaining WHY each clause is risky and proposing specific alternative language, not just flagging
- Mandatory legal disclaimer on every output keeps expectations honest
- One-line install/uninstall scripts with minimal dependencies (nothing beyond Python/reportlab unless you want PDFs)
- Not a lawyer and explicitly not legal advice — every output demands licensed-attorney review before reliance
- Deeply tied to Claude Code-specific mechanics: slash commands, parallel subagents, Read/WebFetch tool references — porting to other clients requires rewriting
- No test suite, benchmarks, or accuracy evidence; analysis quality is unverifiable from the source
- No license declared, so commercial use terms must be confirmed with the author
- The repo description mixes in Skool community marketing, which should be discounted when evaluating capability claims
- README cost-comparison figures (e.g., $300–500/hr attorney rates) are unsourced
How do you install this skill?
Run the repo's one-line installer: curl -fsSL https://raw.githubusercontent.com/zubair-trabzada/ai-legal-claude/main/install.sh | bash — installs all 14 skills, 5 agents, and the PDF scripts at once. PDF reporting requires Python 3.8+ and reportlab (pip3 install reportlab). Uninstall via curl -fsSL .../uninstall.sh | bash or a local ./uninstall.sh. The source does not document exactly where the installer places files.
How do you use this skill?
In Claude Code, type /legal to bring up the command menu, then use commands in their documented form, e.g.: /legal review my-contract.pdf (full review), /legal nda mutual NDA with client Acme (NDA generation), /legal risks <file> (deep risk analysis). Contracts can be supplied by file path, pasted text, or URL; if omitted, the assistant asks for one. The README also describes passing a website URL for terms generation and compliance analysis.
How does this skill compare with similar options?
The source names no competitors; the repo only positions itself within the author's Claude Code Skills Series alongside an AI Marketing Suite and AI Sales Team (unrelated to legal work). No reliable alternative is named — evaluate against human attorney first-pass review or paid contract-review services on your own.