agentleFS
zripticzriptic/roclicks-skills/roblox-game-pipeline/SKILL.md

roblox-game-pipeline

Skill2 starsChanged 29 days agoMITView on GitHub

The end-to-end pipeline for making a Roblox game with your AI agent, from idea to launch and live updates. It runs phases in order (concept and study, design and economy sim, world, models, systems code, UI, VFX, audio, QA, thumbnails and publishing, analytics and live ops) and calls the right skill and tools for each, with standing rules and measured Studio gotchas. Use whenever the developer wants to make, remake, port or clone a Roblox game, start a new game, continue or finish one, plan a game, "do the whole thing", or asks what's next on a game. For one narrow task (just a UI, just an effect) use that skill directly.

The file

We split this skill into the rules it gives an agent

2 rules. About 0 are common, 0 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Find the Studio: use one Studio MCP server for all edits (this skill was written against the community robloxstudio MCP server.

    line 34
  2. Very specific:

    Record: every decision, id and lesson: GAME.md for this game, memory for cross-game facts, PB for generic lessons (no game names or secrets …

    line 103

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Roblox game pipeline

One game, start to finish, with the developer approving each gate. This skill is the conductor: each phase names the skill or tool that does the work, what that phase must produce, and how to know it is done. Do not rewrite what those skills already say; load them when their phase starts.

SK = this skill folder. PB = an optional playbook of your own: a folder (<playbook>/) of generic lessons sorted into topic folders such as game-design/, building/, engineering/ and thumbnails/. If you don't have one yet, skip the PB pointers below and start one as lessons come up.

Companion skills named below (studify, roblox-ui-maker, roblox-vfx, roblox-game-score, claude-music, a design-advice skill, jest-luau, scrapling, agent-browser) and the meshlab toolkit (github.com/zriptic/roclicks-skills, folder meshlab/) are optional. If one isn't installed, do that phase directly with the same "produces" and gate.

Before anything

  1. Find or create the game folder: <projects>/<slug>/. Copy SK/templates/GAME.md into it as GAME.md. That file is the single source of truth: phase status, decisions, place and universe ids, asset ids, open questions. Update it at the end of every phase and whenever a decision lands. A new session on this game starts by reading GAME.md.
  2. Read memory for this game and the developer's standing rules. The rules summary is in SK/references/rules.md; the measured engine and tool facts are in SK/references/gotchas.md. Read both before the first Studio call.
  3. Find the Studio: use one Studio MCP server for all edits (this skill was written against the community robloxstudio MCP server; the official built-in Roblox_Studio server is used only for upload_image). Check which place is open before every write: there are often several open places with the same default name.
  4. Plan the phases you will run for this request and say them in one message: phase, what it produces, which skill. Skip phases the request does not need. A remake of an existing game starts at phase 0 with "study"; a brand-new idea starts at phase 0 with "concept".

The phases

#PhaseProducesSkill / tool
0Concept / studyGAME.md pitch + reference breakdowna design-advice skill, SK/references/study-a-game.md, scrapling / agent-browser for research
1Design + economyPLAN.md: core loop, progression, monetization, numbers; a pacing simdesign-advice skill, PB/game-design/*, a sim/ script (Python or luau)
2WorldMap in Studio, stud style, every visible piece a meshlab modelmeshlab (SK/references/meshlab.md); layout code places baked models; PB/building/* for layout ideas
3ModelsEvery prop, building, vehicle and piece of character gear as meshlab models, baked to MeshDatameshlab (SK/references/meshlab.md, parallel modelling agents); SDF sculpt for creatures; Meshy/Blender only if the developer asks (SK/references/models.md)
4SystemsServer/client Luau: data, remotes, loop, productsPB/engineering/*, jest-luau for tests, luau CLI for pure logic
5UIEvery screen, HTML first, then a real ScreenGuiroblox-ui-maker (always; read its LESSONS.md)
6VFXParticle effects for every key momentroblox-vfx (read its LESSONS.md)
7AudioMusic cues + SFX, SoundGroupsroblox-game-score, claude-music*; keep your verified reusable UI sound ids as described in SK/references/rules.md
8QADefect list fixed; screenshots; loop timingsSK/references/qa.md
9LaunchThumbnails, icon, products, published place, analyticsSK/references/launch.md
10Live opsAnalytics read, next update plannedSK/references/launch.md (analytics), design-advice skill

Phases 2-7 can overlap once phase 1 is approved: a world and its UI do not wait for each other. Systems (4) gate QA (8).

Phase gates (what "done" means)

  • 0 → 1: the developer agrees the pitch and the core loop in one sentence each.
  • 1 → build: PLAN.md approved by the developer; the sim shows the pacing they want (first upgrade under a minute, first rebirth in the target window, no dead stretch). Numbers come from the sim, not from vibes.
  • Each build phase: verified by measurement and screenshots, not by "it compiled" (see qa.md). Show the developer screenshots, ask, and write the feedback into the owning skill's LESSONS.md (UI, VFX) or into memory.
  • 8 → 9: the loop plays start to finish in a playtest with no errors in get_runtime_logs, mobile layout checked, 12+ player-height screenshots reviewed.
  • 9: nothing goes public without the developer's explicit yes (publishing, uploads, dev products, price changes).

Running phases in parallel (lanes)

For a big push, use background subagents as lanes (Agent tool), one per independent phase: world, UI, VFX, audio. Rules and ready-made lane prompts are in SK/references/lanes.md. Studio is a single shared resource: only one lane writes to a given place at a time; the others prepare files (HTML UI, textures, models, music) and hand them back for import.

Always

  • Stud style by default unless the brief or reference says otherwise.
  • Particles, not 3D objects, for effects.
  • HTML-first UI, icons only from the project's chosen icon pack, never emojis.
  • meshlab builds everything visible (the developer's rule): world, props, vehicles, gear. Parts only for invisible colliders, triggers, seats and spawns. Bake to MeshData, and the client rebuilds + ApplyMeshes (SK/references/meshlab.md). No free models.
  • Verify yourself with screenshots and measurements before showing the developer.
  • Ask before spending (Meshy credits, uploads, Robux-priced products) and before anything public.
  • Push Luau through .rbxmx + import_rbxm, never set_script_source for anything with \n in strings (gotchas.md).
  • Record every decision, id and lesson: GAME.md for this game, memory for cross-game facts, PB for generic lessons (no game names or secrets there).

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.