Everything you can do with Darkroom's Claude Code setup. You don't need to memorize this — just describe what you want and Claude will invoke the right skill.
# Install (one command)
bash <(curl -fsSL https://raw.githubusercontent.com/darkroomengineering/cc-settings/main/setup.sh)
# Update
cd ~/.claude/cc-settings && git pull && bash setup.shAfter install, start Claude Code in any project. The setup loads automatically.
Say: "fix the login redirect" or "this is broken" or "debug the auth error"
Triggers /fix — spawns explore → tester → implementer → reviewer agents.
Say: "build a user dashboard" or "add coupon validation"
Triggers /build — research (GO/NO-GO gate) → plan → scaffold → implement → test.
Say: "ship it" or "create a PR"
Triggers /ship — type check → build → test → lint → web quality gate → review → commit → PR.
Say: "review my changes" or "check this before merge"
Triggers /review — reviews against Darkroom standards (TypeScript, React, a11y, performance).
Say: "refactor the payment logic" or "clean up this module"
Triggers /refactor — explore → plan → implement → test → review. Preserves behavior.
Say: "how does auth work?" or "where is the routing logic?"
Triggers /explore — read-only investigation, returns file locations and summaries.
Say: "compare Zustand vs Jotai" or "which approach is better?"
Triggers /f-thread — spawns parallel oracle agents, builds weighted scoring matrix, outputs ADR.
Say: "write a PRD for the notification system"
Triggers /prd — clarifying questions → scope → user stories → task breakdown → execution plan.
Say: "what should I use for state management?" or "advice on caching strategy"
Triggers /ask — delegates to the oracle agent for evidence-based guidance.
Say: "help me figure out what we need"
Triggers /discovery — structured interview to turn vague ideas into clear requirements.
Say: "what could go wrong with this approach?"
Triggers /premortem — analyzes failure modes before they happen.
Say: "double check this is correct" or "verify the auth logic" or "prove it"
Triggers /verify — three-agent adversarial pattern: finder (finds issues) → adversary (disproves them) → referee (judges). For high-stakes code.
Say: "test the payment module" or "add test coverage"
Triggers /test — delegates to tester agent for writing and running tests.
Say: "security review" or "check for vulnerabilities"
Delegates to the security-reviewer agent — OWASP checks, secret scanning, auth audit.
Say: "QA check the homepage" or "does this look right?"
Triggers /qa — screenshot + accessibility snapshot + structured visual review (layout, typography, contrast, hierarchy, a11y).
Say: "compare to the figma design" or "design fidelity check"
Triggers /figma — connects to Figma desktop, screenshots the design, compares against implementation.
Say: "take a screenshot" or "what does the page look like?"
Triggers /debug — pinchtab browser automation for visual debugging.
Say: "how's my context?" or "context usage"
Triggers /context. Watch the statusline:
Opus 4.6 | my-project | main*↑ | ▊░░░░░░░░░ 8% (84k/200K)
| Usage | Action |
|---|---|
| 70-79% | Consider wrapping up or handing off |
| 80-89% | Start wrapping up |
| 90%+ | Run /create-handoff now |
Say: "save a checkpoint" or "checkpoint before this refactor"
Triggers /checkpoint — lightweight JSON snapshot of task state. Used mid-session for quicksaves.
Say: "done for today" or "save state"
Triggers /create-handoff — full markdown session transfer with decisions, files, next steps. Syncs with GitHub Issues if branch is linked.
Say: "resume" or "continue where we left off"
Triggers /resume-handoff — loads handoff + checks linked GitHub Issue for shared context.
| Checkpoint | Handoff | |
|---|---|---|
| Format | JSON | Markdown |
| Weight | Lightweight quicksave | Full session transfer |
| When | Mid-session, before risky ops | End of session, before compaction |
| GitHub sync | No | Yes |
Say: "create a Button component"
Triggers /component — scaffolds components/button/index.tsx + button.module.css with Darkroom conventions.
Say: "create a useLocalStorage hook"
Triggers /hook — scaffolds lib/hooks/use-local-storage.ts with typed options/return interfaces.
Say: "start a new project"
Triggers /init — clones Satus starter template with full Darkroom stack.
Say: "generate a type scale" or "create a color palette"
Triggers /design-tokens — generates tokens as Tailwind config or CSS custom properties.
Say: "set up Lenis" or "add smooth scrolling"
Triggers /lenis — installs and configures Lenis smooth scroll.
Say: "clean up our rules" or "consolidate" or "spa day"
Triggers /consolidate — audits rules, skills, and learnings for contradictions, overlap, and bloat. Merges and prunes.
Say: "remember this" or happens automatically after non-obvious fixes.
Triggers /learn — stores learnings that persist across sessions. Supports --shared for team knowledge via GitHub Projects.
Say: "what's the latest version of gsap?" or triggered automatically before any install.
Triggers /versions — checks bun info <package> before installing.
Say: "how do I use GSAP ScrollTrigger?" or triggered automatically before implementing with a library.
Triggers /docs — fetches current documentation via Context7 MCP. Never codes from memory.
Say: "this is going to be a long task" or "overnight refactor"
Triggers /l-thread — phased execution with automatic checkpoints, verification stack, and recovery from interruption.
Say: "use teams" or "split this across parallel agents"
Triggers /teams — coordinates multiple independent Claude Code instances for true parallelism. Use when 3+ independent workstreams.
Say: "orchestrate this" or "coordinate all the agents"
Triggers /orchestrate — delegates to Maestro for multi-agent coordination.
Say: "think harder" or "quick fix" or "max effort"
Triggers /effort — adjusts reasoning depth. Levels: low, medium, high (default), max.
Activate specialized workflows in settings.json:
| Profile | For |
|---|---|
maestro |
Full orchestration mode |
nextjs |
Next.js web apps |
react-native |
Expo mobile apps |
tauri |
Tauri desktop apps |
webgl |
3D web (R3F, Three.js, GSAP) |
Say: "what am I working on?" or "update the issue"
Triggers /project — reads/updates linked GitHub Issues. Auto-detects from branch name (e.g., feat/123-description).
Say: "who calls this function?" or "find the auth implementation"
Triggers /tldr — token-efficient codebase analysis. 95% fewer tokens than reading files. Semantic search, impact analysis, call graphs, dead code detection.
These are enforced automatically — no skill needed:
- 2-iteration limit — fails twice? Stop, pivot, present alternatives
- Bug fix scope — only touch files related to the bug
- Pre-commit verification — tsc + build + tests must pass before any commit
- Post-compaction recovery — re-read plan + active files after compaction
- Neutral exploration — agents investigate without bias toward expected outcomes
- No AI attribution — stealth mode in all commits and PRs
- Never fake measurements — no fabricated Lighthouse/test/build output
| Skill | Triggers On |
|---|---|
fix |
bug, broken, error, not working |
build |
build, create, implement, add feature |
ship |
ship it, create PR, /pr |
review |
review, check, PR, changes |
refactor |
refactor, clean up, reorganize |
test |
test, write tests, coverage |
verify |
verify, double check, prove it, audit |
explore |
how does, where is, find, understand |
docs |
how to use X, X docs, X API |
ask |
advice, guidance, what should I |
discovery |
requirements, scope, figure out |
prd |
PRD, requirements document, product spec |
premortem |
risks, what could go wrong |
f-thread |
compare approaches, architecture decision |
l-thread |
overnight, long running, autonomous task |
orchestrate |
complex task, coordinate |
teams |
parallel agents, split work, fan out |
project |
project status, update the issue |
tldr |
who calls, dependencies, semantic search |
component |
create component, new component |
hook |
create hook, custom hook |
init |
new project, initialize, setup |
design-tokens |
type scale, color palette, spacing system |
lenis |
smooth scroll, lenis |
debug |
screenshot, visual bug |
figma |
compare to design, figma |
qa |
visual check, accessibility, validate |
versions |
package version, before installing |
context |
context window, running out of context |
checkpoint |
save state, save progress |
create-handoff |
done for today, ending session |
resume-handoff |
resume, continue, last session |
effort |
think harder, max effort, quick fix |
learn |
remember this, store learning (also auto-triggers) |
consolidate |
clean up rules, contradictions, spa day |
| Agent | Role | Delegates To |
|---|---|---|
planner |
Task breakdown, architecture | — |
implementer |
Write and edit code | — |
reviewer |
Code review, quality checks | — |
tester |
Write and run tests | — |
scaffolder |
Boilerplate generation | — |
explore |
Read-only codebase navigation | — |
oracle |
Expert Q&A, deep analysis | — |
security-reviewer |
OWASP, secrets, auth audit | — |
deslopper |
Dead code removal, cleanup | scanners (team mode) |
maestro |
Multi-agent orchestration | all of the above |
| Event | What Happens |
|---|---|
| Session start | Load learnings, check TLDR index |
| User prompt | Skill activation, correction detection |
| Pre-tool (Bash) | Safety net, pre-commit TSC, docs check before install |
| Pre-tool (Edit) | Stale file detection |
| Post-tool (Write/Edit) | Post-edit validation, async TSC |
| Post-tool (TLDR) | Usage tracking |
| Post-tool (git commit) | Commit sound |
| Tool failure | Failure tracking, error sound |
| Pre-compact | Auto-handoff save |
| Stop | Learning reminder, compact reminder |
| Session end | TLDR stats, auto-handoff |
| Subagent start/stop | Swarm logging |
| Notification | Desktop notification |