Flosters

Legal Research — AI skill for Claude Code

AI community

AI-powered legal research skill for Claude Code.

How to install Legal Research

This entry records only its repository, not the path inside it, so there is no exact command to give. Open Flosters/legal-research and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Legal Research does

AI-powered legal research skill for Claude Code. Jurisdiction-specific, source-verified, anti-hallucination. IRAC/CRAC memos with Bluebook, OSCOLA, McGill & AGLC citations.

Alternatives in AI

Open Science — Open Science Desktop — local-first, model-agnostic AI research workbench for macOS, Windows & Linux 1.5k ★
Firm — Convene your standing AI staff — memos on every beat, a board session without you, minutes with dissent preser 1.3k ★
Open Researcher — 🔥 Visual AI research assistant that displays real-time thinking, provides split-view analysis, and automatic 634 ★

README

legal-research-py

A Claude Code skill for autonomous, multi-jurisdiction legal research. It produces verifiable, citation-backed legal reports by orchestrating a sequence of fresh-context subagents — each handling one phase of the research workflow in its own context window.

Install

npx skills add Flosters/legal-research2

Or clone manually and place the directory in `~/.claude/skills/`.

What it does

Given a legal question, the skill:

Extracts the research scope — jurisdiction, area of law, legal posture, language
Designs a verifiable checklist of legal nodes (issues) and N targeted research queries
Creates a dedicated NotebookLM notebook for the research session
Runs N deep-research queries in parallel (one sub-subagent per query, each with its own temp notebook — eliminates the research status race condition)
Curates sources against a jurisdiction-aware priority ladder, deduplicates, and builds an Evidence Registry
Imports primary sources (T1) with crawlability checks and fallback URL resolution
Spot-checks every T1 source for actual queryability in parallel (one sub-subagent per source)
Runs an IRAC-structured analysis sequence via NotebookLM chat
Verifies every citation against the notebook — marks each as ✓ Verified, ~ Paraphrase — Consistent, or [SECONDARY ONLY]
Cross-examines weak propositions via adversarial prompting
Assembles and writes an HTML report

All phases are stateless across subagents — state passes through a schema-validated `state.json` file in the workspace directory.

Workflow

User query
    │
    ▼
Phase 1 — Scope extraction + checklist + query design  (orchestrator, inline)
Phase 2 — Notebook creation                            (orchestrator, inline)
    │
    ▼  state.json written → autonomous from here
    │
Phase 3   — Parallel deep research (Subagent A)
              └─ N query runners dispatched simultaneously (sub-subagents)
Phase 3.5 — Source cu