Darkroom Engineering

Code cleanup agent that suggests improvements and auto-removes dead code only. Reviews codebase for consolidation opportunities, reports findings for approval. Supports team mode for parallel scanning on large codebases. DELEGATE when user asks: - "Find duplicate code" / "Check for redundancy" / "DRY this up" - "Consolidate similar logic" / "Clean up overlapping code" - "Find dead code" / "Remove unused code" / "Clean up the codebase" - After implementer completes changes (before pushing) - "before push" / "pre-push review" / "final cleanup" RETURNS: Dead code auto-removed, consolidation recommendations for approval

Fast codebase exploration, navigation, and documentation fetching. Read-only research agent. DELEGATE when user asks: - "How does X work?" / "Where is X?" / "Find X" - "What files handle Y?" / "Show me the Z code" - "Map the architecture" / "What's the project structure?" - "Find docs for X" / "How do I use Y library?" / "API reference for Z" - "Explain this API" / "What are the options for X?" - "Why is X done this way?" / "Is X safe to change?" / "What happens when Y?" - "Where should I put Z?" / "What's the best practice for X?" - Expert second opinions, blast-radius questions, evidence-based answers - Any exploration before making changes RETURNS: File locations, architecture maps, dependency graphs, code summaries, documentation, evidence-based answers with file:line citations

Code execution agent. Writes, edits, and tests code based on approved plans. DELEGATE when user asks: - "Implement X" / "Build X" / "Create X" / "Add X" - "Fix this bug" / "Make this change" / "Update the code" - "Apply the plan" / "Execute these changes" - After planner has created a roadmap REQUIRED BRIEFING — every prompt MUST contain actual content, not references: 1. The user's original ask, verbatim — not a paraphrase 2. Exact file paths and line ranges to modify — as a subagent this agent gets only your prompt, with none of the conversation context or files you've already read 3. The specific change to make — paste the planner output or quote the recommended fix line-by-line. Never write "based on findings" or "according to plan" 4. The verification command (e.g. `bun test`, `npm run build`, repro steps) 5. Scope boundary — which adjacent code is OFF-LIMITS Thin prompts ("implement based on plan", "fix the bug", "build it") cause regressions. RETURNS: Working code, test results, implementation status, files created/modified

Multi-agent orchestrator. Coordinates complex tasks across multiple agents in parallel. DELEGATE when user asks: - "Implement full feature X" / "Build entire Y system" - "Refactor X across the codebase" / "Large-scale change" - Any task spanning 3+ agents or requiring coordination - Complex debugging requiring multiple perspectives RETURNS: Orchestration status, synthesized results from sub-agents, progress tracking

Task breakdown and architecture planning. Creates detailed implementation roadmaps. DELEGATE when user asks: - "Plan X" / "How should we approach X?" / "Break down X" - "What's the best way to implement Y?" - "Create a roadmap for Z" / "Design the architecture" - Before any multi-file change or complex feature RETURNS: Numbered task lists, dependency graphs, ADRs, risk assessments, phase breakdowns

Autonomous skill-prompt optimization — Karpathy-style mutate/score/keep loop on SKILL.md. Triggers "autoresearch", "optimize skill", "tune", "evolve" a skill, "prompt optimization".

Build feature/page/integration end-to-end with GO/NO-GO gate. For components use `/component`, hooks `/hook`. Triggers "build a feature/page", "implement feature", "add feature", "build integration".

Sync cc-settings with upstream Claude Code (maintainer) or pull the latest cc-settings (user). Triggers "sync with claude code", "changelog sync", "upstream sync" (sync mode); "update cc-settings", "refresh my install" (update mode).

Mid-task rollback points — save/restore state before risky ops (refactors, migrations, destructive edits). For end-of-session save use `/handoff`. Triggers "checkpoint", "snapshot", "before this risky op", "rollback to", "list checkpoints", pre-refactor save.

Create UI component / widget / reusable piece. Triggers "create component", "new component", "add component", or naming a component (Button, Header, Card).