project-guidelines-marketplace

Project Guidelines Plugin

An agent plugin that generates and enforces project coding guidelines. Write your rules, review your code against them.

Compatible with Claude Code, GitHub Copilot in VS Code, and Copilot CLI.

Quick Start

/guidelines-init

It asks one question: "What rules matter to you?" Write your rules, get a PROJECT_GUIDELINES.md, and start reviewing:

/guidelines-review --pr

That's it. The rest of this README covers the details.

Installation

Marketplace (recommended)

# Add marketplace
/plugin marketplace add xanderscannell/guidelines

# Install plugin
/plugin install project-guidelines

Claude Code CLI:

claude --plugin-dir /path/to/Guidelines

VS Code:

Open the Command Palette (Ctrl+Shift+P) and run:

Chat: Install Plugin From Source

then point to this directory or its Git repository URL.

Two Modes

Simple Mode (default)

A flat list of rules in plain language. No categories, no severity levels, no ceremony. You write rules like:

## Rules

- Make sure everything has test cases
- Don't add any more code to `src/legacy/handler.ts`, refactor elsewhere
- Architectural logic belongs in `src/core/`, not in route handlers
- All API responses must use the `ApiResponse` wrapper

The review checks your code against these rules and tells you what's broken. A rule is a rule — if it's in the file, it matters.

Full Mode

Categorized guidelines with severity levels (error / warning / info), structured interviews, per-category review heuristics, and detailed reports with summary tables. Use this when you want comprehensive, team-wide coding standards.

/guidelines-init --full

See Full Mode Details below.

Upgrading

Start simple, go full when you need it:

/guidelines-init --upgrade

This reads your existing simple rules, maps them into categories, adds severity tags, and runs the full auto-detection and interview flow. Your original rules are preserved.

---

Skills

/guidelines-init

Generates a PROJECT_GUIDELINES.md file.

Argument	Mode
(empty)	Simple — quick scan, one question, flat rule list
--simple	Same as empty
--full	Full — comprehensive scan, 8-category interview, severity tags
--upgrade	Converts existing simple file to full mode

Simple mode workflow:

1. Quick scan to identify tech stack
2. "What rules matter to you?"
3. Generates PROJECT_GUIDELINES.md with a flat ## Rules list

Full mode workflow:

1. Auto-detect conventions from 13+ signal files
2. Structured interview across 8 categories (skips what's already detected)
3. Generate categorized guidelines with severity tags
4. Confirm — summary table, adjust severities, then commit

Edge cases handled:

• Monorepo — Detects workspaces and asks about global vs per-package rules
• Existing guidelines — Asks to merge or replace; warns about mode mismatches
• Empty project — Generates a starter template with TODO markers

/guidelines-review [scope]

Reviews code against your guidelines. Auto-detects simple or full mode from the file.

Argument	What gets reviewed
(empty)	All source files in the project
src/api/	All files in that directory (recursive)
src/api/handler.ts	Just that file
--changed	Files changed since the last commit
--staged	Git-staged files only
--pr	Files changed in the current branch vs main/master

Simple mode report:

# Guidelines Review
**Scope:** --pr | **Files scanned:** 8

## Issues Found

### Don't add any more code to `src/legacy/handler.ts`
- `src/legacy/handler.ts:142` — New function added — Move to `src/orders/`

### All API responses must use the `ApiResponse` wrapper
- `src/api/users.ts:38` — Returns raw object — Wrap with `ApiResponse.success(data)`

## Clean
All other rules passed.

Full mode report includes a summary table with pass/warn/fail counts per category, findings grouped by severity (violations, warnings, suggestions), and recommendations for new guidelines.

After the report, both modes offer:

• Fix issues automatically
• Add new rules / update guidelines
• Save report to docs/reviews/

---

Subagent

reviewer

Internal subagent used by /guidelines-review for parallel processing of large codebases (51+ files). Not invoked directly by users.

• Receives file batches of ~20 files
• Simple mode: receives rules list, returns findings per rule
• Full mode: receives guidelines + severity map + category focus, returns categorized findings with severity levels
• Read-only — never modifies files
• Returns structured JSON for the parent skill to merge into the report

---

Full Mode Details

The Severity System

In full mode, every section of PROJECT_GUIDELINES.md has a severity tag: