Reference Library

Purpose: Base reference files and agents, general, reusable across personal, CPA, and investment projects Version: 1.1.0 Last Updated: 2026-05-16

A centralized reference library for authoritative style, drafting, and writing standards. This repository contains no project-specific content. Individual project repositories extend what they need.

API reference: Not applicable. This is a documentation-only repository with no programmatic interface or callable library.

---

What's Here

legal-style/: Oregon Legal Style and Drafting

Reference files extracted from three authoritative Oregon sources:

Source	Governs
Oregon Appellate Courts Style Manual (2002)	Court filings, briefs, opinions
Oregon Legislative Counsel Drafting Manual	Bill drafting, statutory language
Oregon LC Form and Style Manual	ORS citations, bill formatting, grammar

Start with legal-style/QUICK-START.md for the most common daily lookups.

writing-style/: Writing Style Reference

Quantified writing style profile and AI pattern detection reference. Not project-specific; applies to any professional document context.

• writing-style/style-profile.md: sentence metrics, rhythm targets, stylometry (shipped default; per-person calibration lives locally under config/profiles/)
• writing-style/ai-detection.md: blacklisted terms, structural tells, rewrite examples
• writing-style/punctuation-preferences.md: catalog of person-configurable Tier 3 overrides (em-dash, etc.)

config/: Person × Style Profiles

A V2 layer that lets agents be parameterized by person (the maintainer, a colleague, you) and style context (work email, client memo, statutory drafting, court brief, etc.).

• config/profiles.example.toml: shipped default profile plus six starter styles (general, work-email, internal-memo, client-memo, statutory-drafting, court-brief). Committed to the repo.
• config/profiles.toml: your local active config. Gitignored. Created from the example by scripts/setup.sh on first run; edit it to add people, styles, or cross-scoped overrides.
• config/profiles/{person}/{style}.md: per-person calibration data written by the style-analyzer agent. Gitignored.

The resolver script (scripts/profile_resolver.py) merges base defaults → person → style → cross-scoped overrides → call-time arguments and emits the active profile as JSON. Agents call it at invocation time. See CLAUDE.md for resolution rules.

agents/: Base Agent Definitions

General-purpose Claude agents with no project-specific content baked in. Install them globally with scripts/setup.sh; no per-project setup required.

Agent	Pipeline Position	Purpose
style-analyzer	Pre-pipeline (calibration)	Analyze writing samples, compute stylometry, calibrate voice profile
document-drafter	Pre-pipeline (generation)	Voice-calibrated first drafts from outlines, bullets, or prompts
tone-rewriter	Pre-pipeline (generation)	Transform register for a different audience, preserve factual content
grammar-composition-editor	Stage 1 (editing)	Grammar, composition, plain language, AI-mechanical patterns
document-validator	Stage 2 (editing)	Factual accuracy, assumptions, hallucinations, bias, reasoning errors
writing-style-editor	Stage 3 (editing)	Voice alignment, AI pattern detection, stylometry
audience-reaction-analyzer	Post-pipeline (analysis)	Predict audience comprehension, persuasion, emotional response

New users should run the style-analyzer agent first to calibrate the style profile before using the generators or editing pipeline.

---

How to Use Across Projects

Install the agents (one-time setup)

git clone https://github.com/ByronWilliamsCPA/reference-library.git
cd reference-library
bash scripts/setup.sh

This installs all seven agents to ~/.claude/agents/, making them available in every Claude Code project without touching any project files. If you move or reclone the repository, re-run setup.sh to update the installed paths.

Reference files in prompts or agents

After installation, point to library files using the path where you cloned:

See /path/to/reference-library/legal-style/QUICK-START.md for citation rules.
See /path/to/reference-library/writing-style/style-profile.md for style targets.

Extending agents for a specific project

To add project-specific rules on top of a base agent, create a .claude/agents/ file in your project with the additional rules. The base agent's globally installed version handles the standard pipeline; your project file adds the domain-specific layer.

# In project/.claude/agents/writing-style-editor-legal.md
# Extends the base writing-style-editor with Oregon legal style rules.

## Oregon Legal Style Additions
[project-specific legal style rules]

---

Source PDFs

The three Oregon legal PDFs are stored locally in legal-style/source-pdfs/ but excluded from git (see .gitignore). To re-download:

cd legal-style/source-pdfs
curl -O "https://www.courts.oregon.gov/publications/Documents/UpdatedStyleManual2002.pdf"
curl -O "https://www.oregonlegislature.gov/lc/PDFs/draftingmanual.pdf"
curl -O "https://www.oregonlegislature.gov/lc/PDFs/form-stylemanual.pdf"

To re-extract text from the PDFs:

python scripts/extract_legal_pdfs.py

---

Structure

reference-library/
├── README.md
├── agents/                        # Agent templates; installed globally by setup.sh
│   ├── style-analyzer.md
│   ├── document-drafter.md
│   ├── tone-rewriter.md
│   ├── grammar-composition-editor.md
│   ├── document-validator.md
│   ├── writing-style-editor.md
│   └── audience-reaction-analyzer.md
├── config/                          # V2 person × style profiles
│   ├── profiles.example.toml        # Shipped default (committed)
│   ├── profiles.toml                # Active local config (gitignored)
│   └── profiles/                    # Per-person calibration data (gitignored)
├── legal-style/
│   ├── README.md
│   ├── QUICK-START.md
│   ├── cross-reference.md
│   ├── appellate-style-manual/
│   ├── lc-drafting-manual/
│   └── lc-form-style-manual/
├── writing-style/
│   ├── README.md
│   ├── style-profile.md
│   ├── ai-detection.md
│   ├── tone-voice.md
│   ├── structural-formatting.md
│   ├── plain-language-guide.md
│   ├── logical-fallacies-guide.md
│   ├── transition-words-reference.md
│   ├── punctuation-preferences.md   # Tier 3 override catalog
│   └── grammar-style/
├── samples/
│   ├── before/                   # AI-generated before-state samples for pipeline testing
│   └── after/                    # Post-pipeline output: per-stage transcripts plus recaps
├── docs/
│   ├── architecture/             # ADRs
│   ├── compliance-reports/       # Generated audit reports (git-ignored)
│   ├── superpowers/              # Local working directory (git-ignored)
│   ├── known-vulnerabilities.md  # CVE log and accepted Scorecard exceptions
│   └── reusable-workflow-jobs.yaml
├── LICENSES/                     # REUSE 3.0 licenses (FOUND-025)
└── scripts/
    ├── setup.sh                   # Install agents; bootstrap config/profiles.toml
    ├── profile_resolver.py        # Resolve a person × style profile (V2)
    ├── extract_legal_pdfs.py
    └── generate_before_samples.py