beastmode

Turn Claude Code into a disciplined engineering partner.

Workflow patterns for Claude Code. Opinionated like crystallized engineering lore—patterns that survived contact with reality.

The Workflow

A simple loop that scales from quick fixes to deep feature work:

/prime → /design → /plan → /implement → /release → /retro

Quick fix? Prime, implement, done.

New feature? Design the approach. Plan the tasks. Implement. Release. Learn.

Multi-session project? Each phase writes artifacts to .agents/. Next session, /prime restores context. Pick up where you left off.

Every project? /retro feeds learnings back into your agent instructions. Knowledge compounds automatically.

Core Workflow

Skill	Purpose
/prime	Load project context into session
/design	Think through the approach (auto-triggers research when needed)
/plan	Turn design into implementation tasks (auto-triggers research when needed)
/implement	Execute the plan, write the code
/release	Create changelog and release notes
/retro	Capture learnings for next time

Phase Research: Both /design and /plan can automatically spawn a research phase when they detect keywords like "SOTA", "best practices", or unfamiliar tech. Research findings go to .agents/research/.

Utilities

Skill	Purpose
/bootstrap	Initialize project with .agents/ structure
/research	Explore domain when needed
/status	Track project state and milestones

Use what helps. Skip what doesn't.

Install

/plugin marketplace add bugroger/overrides-marketplace
/plugin install beastmode@overrides-marketplace

Then initialize your project:

/bootstrap

The .agents/ Folder

All project context lives here:

.agents/
├── CLAUDE.md           # Project brain (<200 lines)
├── prime/              # Context loaded every session
│   ├── META.md         # How to maintain these docs
│   ├── AGENTS.md       # Multi-agent coordination
│   ├── STACK.md        # Tech stack
│   ├── STRUCTURE.md    # Directory layout
│   ├── CONVENTIONS.md  # Naming patterns
│   ├── ARCHITECTURE.md # System design
│   └── TESTING.md      # Test strategy
├── research/           # Domain exploration
├── design/             # Feature brainstorms, specs
├── plan/               # Implementation tasks
├── status/             # Current state
└── release/            # Changelogs

Your root CLAUDE.md imports: @.agents/CLAUDE.md

Why This Works

Proven workflow. Design before code. Plan before build. Old wisdom, applied to AI coding.

Context survives sessions. Each phase writes artifacts. Next session reads them. No lost work. No repeated explanations.

Scales to complexity. Trivial change? Skip to build. Complex feature? Run every phase. The workflow adapts.

Knowledge compounds. After each implementation, /retro updates your agent instructions. Patterns that work get encoded. Mistakes get documented. Over time, Claude gets smarter about your project—automatically.

No ceremony. No sprint planning. No story points. No standups. Just you, Claude, and the work.

File Naming

• UPPERCASE.md — Meta files with fixed structure (CLAUDE.md, STATUS.md)
• lowercase.md — Content files, flexible structure

Status

Skill	Ready
/prime	✅
/design	✅
/plan	✅
/implement	🚧
/release	🚧
/retro	✅
/bootstrap	✅
/research	🚧
/status	🚧

Credits

• obra/superpowers — Jesse Vincent's original disciplined workflow system
• gsd-build/get-shit-done — Context engineering pioneer

License

MIT