Skills System

Skills are the fundamental building block of UK Legal Skills. Each skill is a self-contained markdown prompt that instructs an agent host how to analyse a specific type of legal document or perform a specific legal task.

Skill File Structure

Every skill lives in its own directory under skills/. The full register is 38 skills — the canonical source is registry/skill-registry.json, enforced by scripts/audit_skills.mjs.

Mermaid source — the diagram above is generated from this

flowchart TB
  REG["<strong>skill-registry.json</strong><br/>38 skills · 12 agents · 11 categories"]
  N["I · legal/SKILL.md (CLI router)"]
  S["II · doc-site (public docs, auto-generated)"]
  W["III · scripts/audit_skills.mjs (parity guardrail)"]
  REG --- N
  REG --- S
  REG --- W

Edit one — the audit enforces the other three.

skills/
├── legal-review/SKILL.md            ← orchestrator (5 parallel agents)
├── legal-employment/SKILL.md        ← orchestrator (4 parallel agents)
├── legal-corporate/SKILL.md         ← orchestrator (3 parallel agents)
└── legal-*/SKILL.md                 ← 35 single-agent or utility skills

For the complete, up-to-date list of skills, run scripts/audit_skills.mjs or read registry/skill-registry.json. The list is intentionally not duplicated here so it cannot drift.

Anatomy of a SKILL.md

Each SKILL.md file contains these sections:

1. YAML Frontmatter

---
name: Contract Review
description: Full contract review with 5 parallel agents
---

2. Jurisdiction Pin

Every skill explicitly states: England & Wales only. This prevents the AI from drifting into Scottish or Northern Irish law.

3. Disclaimer Block

The standard AI-generated analysis disclaimer appears in every skill, ensuring it is included in all output.

4. Purpose Statement

What the skill does, what document types it handles, and what the user can expect.

5. Input Handling

Skills accept input in three ways:

Method	How It Works
File path	The Read tool reads the file directly from disk
Pasted text	User pastes the document content into the chat
URL	WebFetch retrieves the document from a URL

TIP

This input trichotomy is consistent across all skills. When editing ingestion phases, always preserve all three options.

6. Analysis Framework

The core of the skill -- structured phases that guide the model through the analysis. For example:

• Phase 1: Ingestion -- classify the document type, extract metadata
• Phase 2: Analysis -- clause-by-clause review, risk scoring, compliance checks
• Phase 3: Synthesis -- aggregate findings, calculate scores, generate recommendations

7. Output Format Template

A detailed template specifying the exact structure of the output, including section headings, table formats, scoring displays, and recommendation layouts.

8. File Naming Convention

The pattern for saving the output file:

CONTRACT-REVIEW-[name]-[YYYY-MM-DD].md

The Orchestrator

The main orchestrator at legal/SKILL.md is the entry point for all /legal commands. It contains:

1. Command menu -- displayed when users type /legal without arguments
2. Routing table -- maps each command to its skill directory
3. Input handling rules -- the three-way input trichotomy
4. File naming conventions -- standardised output file names
5. Disclaimer -- the standard disclaimer text
6. Tone and style guide -- professional but accessible, with risk indicators

Key Design Principles

Self-Contained

Each skill is a complete, standalone prompt. There is no shared library, no imported modules, no runtime code. You can read a single SKILL.md file and understand everything about that skill.

No Runtime Dependencies

Skills are pure markdown. They require only a compatible agent host to run -- no package installation, no build step, no configuration.

Consistent Structure

All skills follow the same structural pattern, making them predictable to read, edit, and extend.

Adding a New Skill

To add a new /legal command:

1. Create the directory: skills/legal-yourskill/
2. Write SKILL.md following the standard structure
3. Update the routing table in legal/SKILL.md
4. Run ./install.sh to deploy

See the Adding Skills guide for detailed instructions.