Not a chatbot on the side. A teammate inside the pipeline.
Most studios bolt AI onto the edges — a browser tab you paste things into. We built it into the middle. Claude at Lasting Light can open our Unreal project, compile it, run it, look at the scene, and fix what it finds. It also drafts your docs, digests your meetings, and crunches your spreadsheets. One teammate, every desk.
Engineers, designers, artists, QA, production — Claude adapts to the seat it's sitting in. Same teammate, different playbook.
Our shared skill library encodes how Lasting Light builds — exact commands, conventions, and checklists. Browse the full library →
Claude proves its work — builds compile, tests pass, the scene looks right in the editor — before anything is called finished.
Four layers. Learn them once and everything makes sense.
When someone says "ask Claude to do it," this is what's actually happening under the hood. You don't need to memorize any of it — but knowing the layers tells you why Claude here is different from ChatGPT in a browser tab.
ClaudeThe teammate
The intelligence itself. You talk to it in plain English — in Claude Code (terminal or desktop app) when you're working in the project, or the Claude app for everything else. It reads files, writes code, runs commands, browses docs, and works in loops: try, check, fix, repeat. Treat it like a sharp new hire: give it context, tell it what "done" looks like, and check its work at first.
SkillsThe playbooks
Our team's how-to guides, written for Claude instead of humans. A skill packages the exact way Lasting Light does something — the real build commands, our C++ conventions, our asset naming rules, our review checklists. When you invoke one (or Claude picks it up automatically), Claude stops guessing and follows the studio's way. Each department page on this site walks through the skills that matter for your seat.
Unreal MCPThe hands
A live connection between Claude and the Unreal editor. Through the MCP, Claude can spawn and arrange actors, query the scene, read and edit Blueprints, author materials, script Sequencer, run console commands, and take a look at what's actually on screen. Skills are the playbooks; the MCP is the hands that execute them. This is how "move all the torches in this room up 20 units and warm the light color" becomes a sentence instead of an afternoon.
Epic ships this as a first-party plugin now, so it's a one-line install rather than a science project — and the same idea extends to our depot, our crash dashboard, and our tracker. See every connection →
Group KnowledgeThe memory
What one of us teaches Claude, all of us get. Skills live in a shared, versioned library — not on anyone's laptop. When you hit a weird cook failure and figure out the fix, the move isn't "tell the channel," it's "add it to the skill." See exactly how → Next week, everyone's Claude already knows. This is how the studio compounds: our playbooks get sharper every single week.
The library is a real installable package, not a folder people are asked to remember — it updates on every machine on its own. How that's wired →
Find your seat. See what changes.
Each page below is written for one department: what Claude does for you specifically, the skills you'll lean on, real prompts to try on day one, and how the Unreal MCP shows up in your work.
How we work with Claude here.
Six habits. They're the difference between an AI that saves you hours and one that wastes them.
Do
- Give context like you would a new hire. "The pickup respawn logic in BP_PickupBase feels wrong after the refactor" beats "fix the pickups."
- Say what "done" looks like. "Done means the build compiles and the automation tests pass" turns Claude into a self-checking loop.
- Ask it to verify. End Unreal tasks with "prove it in the editor" — Claude will use the MCP to run the map and check its own work.
- Use the skills. If a skill exists for the job, invoke it. That's the studio's accumulated judgment — don't make Claude re-derive it.
- Feed lessons back. Solved something painful? Ask Claude to update the relevant skill and open a PR to the shared library — or use the Contribute page.
- Ask "why," not just "do." Claude is an excellent teacher. "Explain what this replication code does like I'm a designer" is a legit use of its time.
Don't
- Don't ship unreviewed work. Claude's output goes through the same review as anyone's — PRs for code, checks for content. It's a teammate, not an oracle.
- Don't paste secrets. No credentials, keys, or partner-confidential docs into any AI tool that isn't our sanctioned setup. When unsure, ask first.
- Don't fight the editor with it open. If Claude is driving the editor through the MCP, don't edit the same assets simultaneously — same rule as any teammate.
- Don't accept "it should work." If Claude didn't build it, run it, or test it, it doesn't know. Ask it to check.
- Don't keep a fix to yourself. A workaround that lives in your head dies with your session. Skills or it didn't happen.
- Don't wait for permission to experiment. Local work, feature branches, and scratch maps are yours to play in. That's how you learn what it can do.
Up and running in five steps.
Get access & install
Your Lasting Light Claude account comes from your lead. Then follow the Setup guide → — it has the exact one-line install for Mac and Windows, sign-in, and how to connect the Unreal MCP. About 15 minutes, once. If you work in the Unreal project, pull the repo — the shared skills and MCP config ride along with it.
Open it inside the project
Claude Code is dramatically more useful when it's opened in the project folder — it can see the code, the config, and our skills. For non-project work (docs, email, research), the Claude app is the right tool.
Have a real conversation
First prompt: ask it to explain something you actually work with. You'll immediately feel the difference between a search engine and a teammate that has read the codebase.
Run your first skill
Go to your department page on this site, pick the first skill listed, and try its example prompt. Watch what Claude does — you'll see it follow our playbook step by step instead of improvising.
Teach it one thing
Before your first week is out, find one thing Claude got wrong or didn't know — and feed it back into the group knowledge system. That's the whole flywheel. You'll never look at "somebody should write that down" the same way again.
Free official resources worth your time.
Everything on this site gets you working. These go deeper — and they're all free.
Claude Code in Action
Anthropic's hands-on course on working the way we do here: steering long sessions with plan mode, configuring instructions and permissions, automating repetitive tasks, and verifying unsupervised runs. The single best follow-up to this site — take it your first week.
anthropic.skilljar.com → Course catalogAnthropic Academy
The full free catalog — prompting fundamentals, building with the API, MCP deep dives. Good for going from "I use Claude" to "I build with Claude," at your own pace.
anthropic.skilljar.com → Reference docsClaude Code Docs
The official manual: every command, setting, hook, and MCP option. You won't read it cover to cover — you'll search it the moment you wonder "can it do X?" (It can.)
code.claude.com/docs →Past onboarding. The stuff that compounds.
The site doesn't end at day one. This is where you feed the system back, level up past the basics, and settle the questions everyone quietly has.
What's New
Every change to this site with the date and a link straight to it — new connectors, corrected instructions, new pages. Skim the top entry after any announcement and you're current.
what's new → Install it · works offlineThe Companion App
Everything you reach for after you're up and running — the commands, prompt starters by department, connector installs, quick fixes — installable on your phone or desktop and working with no signal.
get the app → The connector libraryConnections
Every verified way to give Claude hands — the Unreal editor, our depot, crash triage, the tracker, localization. Filterable by department, with a trust level on every card and a real command to paste.
connections → For whoever owns the repoStudio Setup
The eight repo-level changes that make everyone else's Claude better by default: code intelligence, the playbook marketplace, guardrails, log monitors, permissions.
studio setup → One-pager · printableThe Cheat Sheet
Every command, the model habit, the everyone-six skills, quick fixes, and a glossary — on a single printable page. Pin it next to your monitor for week one.
cheatsheet → The flywheel, made easyContribute a Skill
The exact template for writing a skill, a before/after example, and a 30-second form to suggest one or flag a problem — no PR required to start.
contribute → Beyond the basicsWeek Two
Parallel work for big sweeps, when to reach past Sonnet, reviewing Claude's output well, and writing your first skill.
week two → Straight answersFAQ
Does it see code outside this project? What if it's wrong? Can I use my personal account? The questions people don't ask out loud.
faq → 30-second pickerWhich Setup Right Now?
Answer two questions, get the model, effort, and mode to use for the thing you're about to do.
quiz →Take it with you.
The site teaches you. The app is what you keep. Commands, prompt starters for your department, connector installs, and quick fixes — installed on your phone and your computer, working with no signal at all.
Free · no app store · 10 seconds Download the companion app Step-by-step for iPhone, iPad, Android, Windows, and Mac — with pictures of exactly what to tap. Looks like an app on your phone, a dashboard on your desk.