Darkroom Claude Code Configuration

Team-shareable Claude Code settings with auto-orchestration, natural language skills, TLDR-first exploration, and persistent learnings.

Installation

Setup Script (Recommended)

bash <(curl -fsSL https://raw.githubusercontent.com/darkroomengineering/cc-settings/main/setup.sh)

Or clone and run locally:

git clone https://github.com/darkroomengineering/cc-settings.git /tmp/darkroom-claude
bash /tmp/darkroom-claude/setup.sh

Options:

• --minimal - Skip optional tools (agent-browser, tldr)
• --skip-deps - Skip all dependency installation

Alternative: Plugin Install

If you have SSH keys configured for GitHub:

/plugin install darkroomengineering/cc-settings

Or with HTTPS:

/plugin install https://github.com/darkroomengineering/cc-settings

Dependencies

The setup script auto-installs:

• jq - Required for learnings, statusline
• llm-tldr - Semantic code search (95% token savings)
• agent-browser - AI-optimized browser automation

Restart Claude Code to apply changes.

---

Philosophy

You don't need to memorize commands. Just describe what you want naturally.

Claude automatically:

1. Recognizes intent from your message
2. Invokes the appropriate skill
3. Delegates to specialized agents
4. Returns a clean summary

You: "Fix the auth bug"

Claude automatically:
→ Invokes fix skill
→ Spawns explore agent (investigate)
→ Spawns implementer agent (fix)
→ Spawns tester agent (verify)
→ Returns summary

---

What Gets Installed

~/.claude/
├── CLAUDE.md           # Coding standards + orchestration mode
├── settings.json       # Permissions + Hooks + MCP
├── agents/             # 10 specialized agents (TLDR-enforced)
├── skills/             # 23 auto-invocable skills (SKILL.md format)
├── contexts/           # 4 ecosystem contexts (web, webgl, desktop, mobile)
├── rules/              # 7 path-conditioned rule files
├── profiles/           # 4 stack-specific profiles
├── mcp-configs/        # MCP server templates
├── scripts/            # 12 hook scripts
├── lib/                # 5 shared bash utilities
├── learnings/          # Persistent project memory
├── handoffs/           # Session state backups
└── tldr-cache/         # TLDR warm status

---

Core Features

Auto-Orchestration (Maestro Mode)

Claude defaults to orchestration mode—coordinating agents rather than executing directly.

You Say	Claude Does
"Fix the broken login"	explore → implementer → tester
"Build a dashboard"	planner → scaffolder → implementer
"How does auth work?"	explore with TLDR semantic search
"Review my changes"	reviewer agent
"Done for today"	create-handoff (critical)

Skills Auto-Invoke

Skills use the native SKILL.md format with intent-based descriptions. Claude reads these descriptions and automatically invokes the right skill based on your conversation.

Natural Language	Skill Invoked	What Happens
"fix", "broken", "bug"	fix	explore → implement → test
"build", "create", "add feature"	build	plan → scaffold → implement
"how does X work?"	explore	TLDR semantic + context
"refactor", "clean up"	refactor	explore → implement → review
Non-obvious fix discovered	learn	Auto-stores the insight

TLDR-First Exploration

All agents enforce TLDR usage for 95% token savings:

# Instead of reading files...
tldr context functionName --project .   # LLM-ready summary
tldr semantic "authentication flow" .   # Find by meaning
tldr impact functionName .              # Who calls this?
tldr slice file.ts func 42              # What affects line 42?

Agents have "Forbidden" lists that prohibit reading raw files when TLDR would suffice.

Auto-Learning

The learn skill auto-invokes when Claude discovers something worth remembering:

• Non-obvious bug fixes
• Useful patterns
• Gotchas and edge cases
• Architecture decisions

Learnings persist across sessions and are recalled on session start.

Statusline

Custom status bar showing model, directory, git branch with status, and context window usage:

Claude 4.5 Opus | my-project | main✱↑ | ████░░░░░░ 42% (84k/200k)

---

Natural Language Examples

What You Say	What Happens
"Fix the broken login"	fix workflow (explore → tester → implementer)
"Build a user dashboard"	build workflow (planner → scaffolder → implementer)
"How does auth work?"	explore agent with TLDR
"Review my changes"	reviewer agent
"Create a Button component"	scaffolder agent
"What could go wrong?"	premortem analysis
"Done for today"	saves session state
"Resume where we left off"	loads previous handoff

---

Skills Reference

Workflows (Multi-Agent)

Skill	Triggers	Agents
fix	bug, broken, error, not working	explore → tester → implementer
build	build, create, implement, add feature	planner → scaffolder → implementer
refactor	refactor, clean up, reorganize	explore → implementer → reviewer
review	review, check, PR	reviewer
test	test, write tests, coverage	tester
orchestrate	complex task, coordinate	maestro

Research (Forked Context)

Skill	Triggers
explore	how does, where is, find, understand
docs	documentation, how to use, API
ask	advice, guidance, what should I
tldr	who calls, dependencies, semantic search
premortem	risks, what could go wrong
discovery	requirements, scope, figure out

Creation (Direct Output)

Skill	Triggers
component	create component, new component
hook	create hook, custom hook
init	new project, initialize

Session Management

Skill	Triggers
learn	AUTO: after discoveries
context	context window, running out
create-handoff	done for today, save state
resume-handoff	resume, continue, last session

---

Agents

All agents are TLDR-enforced with mandatory commands and forbidden actions.

Agent	Purpose	TLDR Commands
explore	Codebase navigation + docs	semantic, arch, context
oracle	Expert Q&A	context, semantic, impact, slice
planner	Task breakdown	arch, impact, calls
implementer	Code execution	context, impact, slice
reviewer	Code review	impact, context, change-impact
tester	Testing	change-impact, impact, context
scaffolder	Boilerplate	semantic, structure, context
maestro	Multi-agent coordination	All commands
security-reviewer	Security audit	semantic, slice, dfg, impact

---

Rules (Path-Conditioned)

Rules load automatically based on file context:

Rule	Loaded When
react.md	Working with .tsx, .jsx, components/
typescript.md	Working with .ts, .tsx
style.md	Working with CSS, SCSS, styled components
accessibility.md	Working with UI components
security.md	Working with API routes, lib code, env files
performance.md	Working with app code or components
git.md	Always loaded

---

Ecosystem Contexts

Switch contexts for different platforms:

/context web      # Default - Next.js, React, Tailwind
/context webgl    # R3F, Three.js, GSAP, shaders
/context desktop  # Tauri (Rust + Web)
/context mobile   # Expo (React Native)

---

MCP Servers

Server	Purpose	Status
context7	Up-to-date library docs	Auto-configured
Sanity	CMS operations (GROQ, documents)	OAuth on first use
tldr	Semantic search, impact analysis	Auto-installed

---

Native Hooks

Event	Action
SessionStart	Recalls learnings, auto-warms TLDR
SubagentStart/Stop	Swarm logging
PostToolUse	TLDR tracking (async)
PreCompact / SessionEnd	Auto-handoff

---

Quick Reference

Want To...	Just Say...
Understand code	"How does X work?"
Build feature	"Build a..."
Fix bug	"Fix the..."
Review code	"Review my changes"
Find callers	"Who calls X?"
End session	"Done for today"
Resume	"Resume where we left off"
Debug visually	"Take a screenshot of..."
Security audit	"Check for security issues"

---

Documentation

Doc	Content
USAGE.md	Daily workflow, examples
skills/README.md	Skill system details
hooks/README.md	Native hooks configuration
rules/README.md	Path-conditioned rules
mcp-configs/README.md	MCP server setup
contexts/README.md	Ecosystem contexts

---

Platform Support

• macOS - Native (Homebrew for dependencies)
• Linux - Native (apt, dnf, pacman)
• Windows - Via Git Bash, WSL, or MSYS2

---

Contributing

1. Edit your local ~/.claude/ config
2. Test with Claude Code
3. Copy to this repo and submit PR