The bundled toolkit

The toolkit is the curated Singapore instance of the architecture. A single command assembles it into a project folder for an AI host such as Claude Code or Codex: every skill is a snapshot pinned to an exact version, with an integrity record so that later changes can be detected.

The skills

Skill	Layer	What it does	What it never decides
Toolkit guide	Connects all three	Explains which layer does what, lists the installed skills and contracts, resolves each workflow step to a capability, and releases signed-off workflows	Anything a person or another skill decides
Process modelling and runner	3	Models a lawyer's process, states each step's need, and runs released workflows with recorded state and human gates	Legal criteria the owner has not supplied
Scrum delivery and dashboard	2	Turns a signed-off design into a backlog, acceptance criteria and Sprints, and reports progress	A matter outcome
Singapore legal research	1	Researches and cite-checks Singapore law from primary sources, with a private evidence ledger	Whether research may be relied upon
Singapore deadlines	1	Computes procedural, limitation and notice dates from an explicitly chosen rule, with a full derivation trace	Which rule applies
Source-span extraction	1	Extracts fields from documents, binding every value to its exact source text	What a clause means in law
Singapore clause drafting	1	Drafts a clause and drafting note from supplied research only, and captures approved clauses as precedents	Whether the clause is approved
PDPA assessment	1	Assesses an activity against each obligation of the Personal Data Protection Act, from provision text retrieved for that assessment	Whether to proceed or notify anyone

A released workflow adds one more skill: a launcher named after the workflow, through which a lawyer starts a matter.

Contracts: what each capability must prove

Each Layer 1 capability implements a contract: the inputs it needs, the output it returns, and the consistency rules a reviewer relies on. Any skill that meets a contract can serve the same workflow step.

Contract	Refuses, for example
Research	Claiming completed research without a passing evidence check and a supporting authority
Cite-check	Reporting every citation as supported while one is unverifiable
Drafting	A drafting note citing an authority, pinpoint or statute that is not in the supplied research
Review	A gap without a recommendation, or a checklist item left unaddressed
Extraction	A value marked as found without the source text it came from
Computation	A date from an unverified rule or holiday year without saying so
Capture	Retaining a clause as a precedent without the approver and their approval

These rules test whether a record is honest and complete. They do not test whether the legal analysis is right; that remains the reviewing lawyer's judgement.

Components and installation

Each skill lives in its own repository and is pinned by exact commit in the toolkit's lock file. Repositories marked private are available on request while they are reviewed.

Repository	Provides	Access
legal-engineering-defined	This guide, the installer, the contracts, bindings and toolkit guide skill	Private
legal-processmodelling-meta-skill	Process modelling and the workflow runner (Layer 3)	Private
scrum-skill	Scrum delivery and dashboard (Layer 2)	Public
sg-legalresearch-skill	Singapore legal research and cite-check	Public
sg-deadline	Singapore deadlines	Public
legal-extraction-skill	Source-span extraction	Public
sg-drafting-skill	Singapore clause drafting and precedent capture	Private
sg-pdpa-skill	PDPA assessment	Private

With Python 3.11 or later, Git, and access to the private repositories:

git clone https://github.com/kevanwee/legal-engineering-defined.git
cd legal-engineering-defined
python -m pip install -e .
legal-eng init --workspace ../my-legal-workspace
legal-eng doctor --workspace ../my-legal-workspace

The workspace folder is opened in Claude Code or Codex. Its CLAUDE.md and AGENTS.md direct each session to start from the toolkit guide skill. The operator describes a process requiring improvement or requests a released workflow by name. init also prepares skill packages for ChatGPT and the Claude app, installed through each product's skills interface.

Beyond the bundled skills

Layer 1 is open. Other skill collections can serve as domain capabilities where they fit a need, for example the open-source skills published by LegalQuants and the legal plugin in Anthropic's knowledge-work plugins, both under the Apache 2.0 licence. A capability from outside the toolkit can be wrapped for a workflow step, with the record noting that its version is not pinned by the toolkit.

Adding a capability to the toolkit itself means implementing a contract, publishing the skill, and pinning a binding to its exact version. A new version of any skill means new versions of the workflows that use it; released workflows never change underneath the matters running on them.

Where each part stands

Component	Status
Installation and integrity checks	Tested; a fresh assembly from the published sources passes its checks
Research and cite-check method	In use on synthetic and saved sources; live statutory access depends on the host's browser and publisher terms
Deadlines	Engine tested; every bundled rule and holiday year is still marked unverified until checked against primary sources
Extraction	Tested; its checks are mechanical and make no legal claims
Drafting	Checks tested with synthetic fixtures; not yet evaluated on a real clause by a reviewing lawyer
PDPA assessment	Checks tested with synthetic fixtures; the obligation map has not yet been reviewed by a data protection lawyer
Release and launcher	Tested, including refusal of unsigned, pending or altered releases
A complete live matter	Not yet run

The reference chapter explains the terms used here and the boundary of the current implementation.

Continue with Building the practice. The next part explains the methods behind the journey in depth, starting with how LegalTech, Legal Operations and Legal Engineering differ.