Disciplined Process Plugin
A rigorous, traceable AI-assisted development workflow plugin for Claude Code.
Why Use This?
AI coding assistants are powerful but can produce inconsistent results. This plugin brings discipline to AI-assisted development by enforcing a proven workflow:
- Specify before you code — Write requirements with traceable IDs before implementation
- Test before you implement — Tests reference specs and must exist before code
- Document decisions — Architecture Decision Records capture the "why"
- Track dependencies — Know what's ready to work on and what's blocked
- Enforce or guide — Strict mode blocks bad commits; guided mode just warns
Inspired by the Rue language development process.
Features
- Specification-first: Write specs before code, with traceable paragraph IDs (
[SPEC-01.03])
- Test-driven: Tests reference specs via
@trace markers, run before implementation
- ADRs: Document architectural decisions systematically
- Task tracking: Dependency-aware work management (Beads default, or GitHub/Linear/Markdown)
- Traceability: Every line links to specs — know why code exists
- Enforcement: Configurable hooks enforce the process (strict), warn (guided), or stay out of the way (minimal)
The Workflow
A 7-phase loop: Orient → Specify → Decide → Test → Implement → Review → Close
Each phase has supporting commands (/dp:task, /dp:spec, /dp:adr, /dp:review). See full documentation for details.
Quick Start
# 1. Add the plugin repository
/plugin marketplace add rand/disciplined-process-plugin
# 2. Install the plugin
/plugin install disciplined-process@disciplined-process-plugin
# 3. Initialize your project (interactive wizard)
/dp:init
# 4. Start working
/dp:task ready
See full documentation for detailed usage.
Installation
Prerequisites
- Claude Code version 2.0.0 or higher
- Git (for Beads task tracker, if used)
Step-by-Step
-
Add the plugin repository:
/plugin marketplace add rand/disciplined-process-plugin
-
Install the plugin:
/plugin install disciplined-process@disciplined-process-plugin
-
Verify installation:
/plugin list
You should see disciplined-process listed.
-
Initialize your project:
/dp:init
The wizard will configure language, task tracking, test frameworks, and enforcement level.
-
Validate setup:
/dp:help # Should show command reference
cat .claude/dp-config.yaml # Should show your configuration
Commands
| Command | Description |
|---|
/dp:init | Initialize project with interactive wizard |
/dp:task | Task tracking (ready, create, show, update, close, discover) |
/dp:spec | Specification management (create, add, coverage) |
/dp:adr | Architecture Decision Records |
/dp:review | Code review checklist |
/dp:verify | Goal-backward verification |
/dp:trace | Traceability validation |
/dp:decompose | Decompose specs into work items |
/dp:progress | Generate progress reports |
/dp:session | Session management |
/dp:migrate | Migrate between task trackers |
/dp:status | Show degradation status |
/dp:health | System health and diagnostics |
/dp:help | Help and workflow reference |
Updating
/plugin marketplace update rand/disciplined-process-plugin
/plugin update disciplined-process@disciplined-process-plugin
Important: After updating, re-merge hooks to update paths:
python3 ~/.claude/scripts/merge-plugin-hooks.py
Hooks
This plugin uses a hybrid hook architecture combining prompt-based hooks (for LLM judgment calls) with compiled Go binaries (for fast, deterministic validation).
Go Validation Binaries (v2.1.0+)
Pre-compiled binaries in bin/ provide fast validation without runtime dependencies:
| Binary | Purpose |
|---|
trace-validator | Recursively discovers test files and validates @trace SPEC-XX.YY markers |
coverage-check | Checks spec coverage against test traces |
adr-validator | Validates ADR format and required sections |
phase-emitter | Emits phase-transition events for cross-plugin coordination |
Binaries are cross-compiled for Linux (amd64/arm64), macOS (amd64/arm64), and Windows (amd64).
Prompt-Based Hooks
Prompt hooks inject context into Claude's reasoning for judgment-based enforcement (e.g., "is this commit ready?" decisions that require understanding intent).
Hook Summary
rand