Core orchestration for disciplined AI-assisted development. Auto-invokes when starting new features, implementing tasks, or working on any development that should follow the spec-first, test-driven process. Triggers on task planning, feature implementation, bug fixes, refactoring work, or when beads/bd tasks are referenced.
From disciplined-processnpx claudepluginhub rand/disciplined-process-plugin --plugin disciplined-processThis skill uses the workspace's default tool permissions.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
Searches prompts.chat for AI prompt templates by keyword or category, retrieves by ID with variable handling, and improves prompts via AI. Use for discovering or enhancing prompts.
Compares coding agents like Claude Code and Aider on custom YAML-defined codebase tasks using git worktrees, measuring pass rate, cost, time, and consistency.
This skill enforces a rigorous, traceable development process inspired by the Rue language project. Every implementation traces back to specifications, every specification has tests, and every decision is recorded.
For any non-trivial work, follow the 7-phase loop:
Orient → Specify → Decide → Test → Implement → Review → Close
See references/workflow.md for the full workflow reference with commands.
For large features (multi-file, multi-task), use these tools to manage complexity:
/dp:decompose): Transforms specs into dependency-aware, context-window-sized work items. Ambiguities become first-class "holes" that block downstream work. Use after the Specify phase for feature-scale work./dp:progress): Generates structured reports from Beads and git state. Reports serve both humans (status updates, action items) and AI agents (context recovery for fresh sessions via docs/progress/latest.md). Use during Close phase or in watch mode for ongoing projects.Before implementing anything:
# Check what's ready to work on
/dp:task ready
# Claim a task
/dp:task update <task-id> --status in_progress
# Review the task context
/dp:task show <task-id>
When you discover related work:
# File discovered issues, linked to parent
/dp:task discover "Found: <issue>" --from <parent-id> -p <priority>
# Run quality gates (tests, lint, build)
# Close the task
/dp:task close <task-id> --reason "Implemented per spec [SPEC-ID]"
# Commit with task reference
git add -A && git commit -m "feat: <description> (<task-id>)"
The disciplined process expects this structure:
project/
├── CLAUDE.md # Project-specific adaptations
├── .claude/
│ └── settings.json # Hooks and permissions
├── .beads/ # Task tracking (or alternative)
├── docs/
│ ├── spec/ # Specifications with paragraph IDs
│ │ ├── 00-overview.md
│ │ ├── 01-<domain>.md
│ │ └── ...
│ ├── adr/ # Architecture Decision Records
│ │ ├── 0001-<decision>.md
│ │ └── template.md
│ └── process/ # Process documentation
│ ├── code-review.md
│ └── workflow.md
├── tests/ # Test suites by type
│ ├── unit/
│ ├── integration/
│ ├── property/
│ └── e2e/
└── src/ # Implementation
This workflow supports three enforcement levels (configured in .claude/settings.json):