workflow

Workflow Skills

A complete feature development workflow for AI coding agents. Structured SDLC phases, scope-based routing, and requirement traceability from PRD through to shipped code.

Works with Claude Code, Cursor, Codex, OpenCode, and Gemini CLI.

Quick Start

Claude Code / Cursor

/plugin install workflow@dirkkok-skills

Codex

git clone https://github.com/dirkkok101/skills.git ~/.codex/workflow-skills
ln -s ~/.codex/workflow-skills/skills ~/.agents/skills/workflow

Gemini CLI

gemini extensions install https://github.com/dirkkok101/skills

OpenCode

See .opencode/plugins/workflow.js.

Initialize Your Project

/workflow:init

Skills

Command	When to Use
/workflow:init	Starting a new project, setting up docs structure
/workflow:research	Need competitive analysis, landscape review, prior art
/workflow:brainstorm	New feature idea, exploring approaches
/workflow:discovery	Complex feature requiring deep domain requirements
/workflow:prd	Writing formal product requirements
/workflow:technical-design	Architecture, API specs, data models
/workflow:plan	Breaking design into implementation tasks
/workflow:beads	Creating work packages from approved plan
/workflow:execute	Implementing from approved beads
/workflow:review	Multi-agent code review after implementation
/workflow:review-prd	Quality gate between PRD and design
/workflow:review-design	Quality gate between design and plan
/workflow:review-plan	Quality gate between plan and beads
/workflow:review-beads	Quality gate between beads and execute
/workflow:review-execute	Post-execution bead satisfaction verification
/workflow:qa	Browser-based QA testing
/workflow:benchmark	Performance benchmarking
/workflow:security-audit	OWASP + STRIDE security audit
/workflow:ship	Create PR with changelog and traceability
/workflow:compound	Capture learnings for future sessions
/workflow:diagnose	Bug investigation and root cause analysis
/workflow:autoresearch	Automated document quality convergence

Pipeline

research ─> brainstorm ─> discovery ─> prd ─> technical-design ─> plan ─> beads ─> execute
(optional)                 (COMP only)

  execute ─> qa ─> benchmark ─> review ─> review-execute ─> security-audit ─> ship ─> compound
             (opt)   (opt)                                      (opt)

Quality gates (review-*) are optional between pipeline stages.

Scope-Based Routing

Brainstorm classifies features and routes to the right pipeline depth:

Scope	Pipeline
BRIEF (0-2 pts)	brainstorm → plan → beads → execute → review → ship
STANDARD (3-4 pts)	brainstorm → prd → technical-design → plan → beads → execute → review → ship
COMPREHENSIVE (5+ pts)	brainstorm → discovery → prd → technical-design → plan → beads → execute → review → ship

Other Entry Points

Scenario	Path
Known requirements	prd → technical-design → plan → beads → execute → review → ship
Technical improvement	brainstorm → technical-design → plan → beads → execute → review → ship
Bug fix	diagnose → fix / beads / brainstorm
Pre-release check	qa → benchmark → review → security-audit → ship

Key Concepts

Traceability Chain — Requirements trace from PRD → Design → Plan → Beads → Execute → Review. Each phase references upstream artifacts by stable FR IDs.

Intent Over Implementation — Beads carry objectives and acceptance criteria, not code. The executing agent writes code from codebase patterns.

Surgical Context — Each bead specifies exactly which files to read. Context resets between beads to prevent drift.

Self-Regulation — Execute and QA track cumulative risk scores. High scores trigger PAUSE or STOP to prevent doing more harm than good.

CONVERGE Mode — Review skills auto-fix mechanical findings and re-review until 0 FAILs or convergence. Judgment calls are escalated to the user.

Stage Gates — Every phase pauses for user approval via AskUserQuestion before proceeding. Five interaction patterns: Decision Gate, Comparison, Batch Review, Guided Review, Combined.

Project Structure After Init

your-project/
├── CLAUDE.md
├── .beads/beads.db
└── docs/
    ├── prd/           # Requirements
    ├── designs/       # Technical design
    ├── plans/         # Implementation plans
    ├── execution/     # Execute manifests
    ├── reviews/       # Review reports
    ├── learnings/     # Compound learnings
    ├── adr/           # Architecture decisions
    ├── patterns/      # Coding patterns
    └── architecture/  # System architecture

Prerequisites

Required: br (beads-rust) — local issue tracking for beads