cut
Skill39k starsChanged 7 days agoMITView on GitHub
Proposes cuts and a polish list to finish the game from its brief and what is built; the person picks each cut. A game to sell also gets a release date and a feature freeze. Use when asked what to cut, how to finish or ship it, or for a polish list.
6 other projects use a version of this.
Adopted by
Steady: 0 new projects took it in the last 4 weeks, none the 4 weeks before.thin data
Most recently taken up by
- clnnn/playbooktheir own version, first seen Nov 8, 2026
- ivala2081/vibe-gametheir own version, first seen Nov 8, 2026
- lunaman81/claude-skillstheir own version, first seen Oct 6, 2026
- fuuuuuuma/premiere-skillstheir own version, first seen Sep 18, 2026
- barefootford/buttercuttheir own version, first seen Sep 14, 2026
- Megaprompting/torque-looptheir own version, first seen Jul 3, 2026
Who's talking about Donchitos/Claude-Code-Game-Studios
- Claudical“그림 파일 하나 없이 아이디어 한 줄로 만든 게임 | Claude Code Game Studios”YouTube · Oct 27, 2026 · 18k views
- Matheus Battisti“Criar jogos com IA ficou ABSURDO! (Claude Code com 49 agentes)”YouTube · Nov 2, 2026 · 5.6k views
- Chrono IT Channel“Letting 49 AI Agents Run a "Game Studio" — Claude Code Game Studios”YouTube · May 30, 2026 · 1.8k views
- AutomateNow“Claude Code Game Studios: Free Repo Turns Claude Code Into a 49-Agent Game Studio #Shorts”YouTube · Oct 30, 2026 · 1.4k views
- SkipCI“Claude Code Game Studios: 49 AI Agents, 74 Commands, Ranked Last in Blind Test”YouTube · Oct 31, 2026 · 1.4k views
The biggest 5 of 35 mentions linking to the repository that our crawl found. Each line is their own headline.
See all 35
Talked about on YouTube and Bluesky (35 times).
On YouTube · 23
- Oct 27, 2026그림 파일 하나 없이 아이디어 한 줄로 만든 게임 | Claude Code Game Studios — Claudical18k views · 9 comments
- Nov 2, 2026Criar jogos com IA ficou ABSURDO! (Claude Code com 49 agentes) — Matheus Battisti - Hora de Codar5.6k views · 8 comments
- May 30, 2026Letting 49 AI Agents Run a "Game Studio" — Claude Code Game Studios — Chrono IT Channel1.8k views · 3 comments
- Oct 30, 2026Claude Code Game Studios: Free Repo Turns Claude Code Into a 49-Agent Game Studio #Shorts — AutomateNow1.4k views · 3 comments
- Oct 31, 2026Claude Code Game Studios: 49 AI Agents, 74 Commands, Ranked Last in Blind Test — SkipCI1.4k views · 4 comments
18 more on YouTube
- Oct 31, 2026Claude Code Game Studios: một phiên làm việc, cả studio game — Anh Nguyen Quoc (C5)813 views
- Oct 28, 2026Turn Claude Code Into a Game Studio (Part 1) — AI → AGI → ASI801 views
- Apr 16, 2026Donchitos/Claude-Code-Game-Studios - Gource visualisation — Gourcer740 views
- Oct 15, 2026Claude Code Game Studios GitHub Overview: How Sequential AI Roles Manage Game Projects — Alex Hitt734 views
- Oct 31, 2026Claude Code Game Studios: one session, a full game studio — GitHub Unboxed609 views
- Oct 28, 20261 Claude Code Chat = A Whole Game Studio 🤯 — Gadget Drop282 views · 1 comment
- Oct 15, 2026Claude-Code-Game-Studios: virtual game studio with a single chat — Next Horizon191 views · 1 comment
- Oct 16, 2026Claude Code Game Studios: 49 AI Agents to Run Your One-Person Game Studio — Not That Number188 views
- Oct 29, 2026Donchitos/Claude-Code-Game-Studios — Turn Claude Code into a full game dev studio — 49 AI agents, 72 — Git Radar104 views
- Nov 1, 2026Make a game solo like a whole studio? 49 agents in Claude Code | GitHub Radar #12 #Shorts — KIEU STUDIO54 views · 1 comment
- Oct 22, 2026Claude Code como estudio de VIDEOJUEGOS 🎮 49 agentes: ¿mito o verdad? (Game Studios) — BarbuPrime IA | Claude Code Vs Codex32 views · 1 comment
- Oct 31, 2026Donchitos/Claude-Code-Game-Studios is trending on GitHub. — Nexbit Studio18 views
- Oct 30, 2026게임 개발 도와주는 AI? Claude Code Game Studios로 할 수 있는 것 — 딴삽스튜디오12 views
- Oct 29, 2026Claude-Code-Game-Studios ⭐ 28601 | +538 today #github #opensource #shell — Git Radar12 views
- Nov 3, 2026Claude Code Just Became a FULL Game Development Studio 🎮🤖 — The Clever Machine10 views
- Nov 2, 2026Claude Code as a 49-Agent Game Studio (and Why More Docs Lost) — GitHub AI Repo10 views
- Oct 31, 2026【AIが49人のゲーム会社に】Claude Code Game Studiosの入れ方と全6ステップ|作らせたいゲーム募集 — にくまるOSS研究所8 views
- Nov 3, 2026【Claude-Code-Game-Studios】49エージェントの"ゲーム開発スタジオ"に1画面ゲームを頼んだらminimalは38分で遊べる物・standardは8ドルで設計書9本 — さつきのOSS研究室4 views · 1 comment
On Bluesky · 12
- Apr 19, 2026【GitHubトレンド調査】 Claude Codeをゲーム開発スタジオに変貌させる「Claude-Code-Game-Studios」が登場。… — @woody-ai-tools.bsky.social10 likes and reposts
- Apr 15, 2026🎉 Celebrating! 🎉 (500+ new stars) 📦 Donchitos / Claude-Code-Game-Studios ⭐ 9,860 (+523) 🗒 Shell Turn Claude Code into a full game dev… — @github-trending.bsky.social4 likes and reposts · 1 reply
- May 19, 2026This is the project in question. github.com/Donchitos/Cl... Most of these projects feel like toys or performance art masquerading as useful… — @impossible.bsky.social3 likes and reposts · 2 replies
- Apr 19, 2026🎉 Celebrating! 🎉 (500+ new stars) 📦 Donchitos / Claude-Code-Game-Studios ⭐ 13,024 (+828) 🗒 Shell Turn Claude Code into a full game dev… — @github-trending.bsky.social3 likes and reposts · 1 reply
- Nov 1, 2026Claude-Code-Game-Studios — Turn Claude Code into a full game dev studio — 49 AI agents, 72 workflow skills, and a complete coordination… — @thestackradar.bsky.social2 likes and reposts
7 more on Bluesky
- Nov 1, 2026An entire game dev studio in one Claude session? 🎮 Claude-Code-Game-Studios turns Claude Code into an AI production studio: 49 agents, 72… — @thestackradar.bsky.social1 likes and repost
- Oct 31, 2026今日のGitHubトレンド Donchitos/Claude-Code-Game-Studios Claude Codeのセッションを擬似的なゲーム開発スタジオとして運用するためのツールです。… — @dailygithubtrends.bsky.social1 likes and repost
- Apr 21, 2026https://github.com/Donchitos/Claude-Code-Game-Studios — @tom-doerr.bsky.social1 likes and repost
- Apr 1, 2026Un studio de 48 agents IA spécialisés dans Claude Code : hiérarchie complète (directors, leads, specialists), workflows automatisés et… — @camilleroux.com1 likes and repost
- Nov 2, 2026💎 Hidden Gem! 💎 (1000+ new stars) 📦 Donchitos / Claude-Code-Game-Studios ⭐ 33,084 (+1,380) 🗒 Shell Turn Claude Code into a full game… — @github-trending.bsky.social0 likes and reposts · 1 reply
- Oct 29, 2026🎉 Celebrating! 🎉 (500+ new stars) 📦 Donchitos / Claude-Code-Game-Studios ⭐ 28,645 (+538) 🗒 Shell Turn Claude Code into a full game dev… — @github-trending.bsky.social0 likes and reposts · 1 reply
- May 19, 2026It was this one: github.com/Donchitos/Cl... — @dphrygian.com0 likes and reposts · 1 reply
Other versions6 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 6
- barefootford/buttercut★ 625same name
- fuuuuuuma/premiere-skills★ 14same name
- Megaprompting/torque-loop★ 10same name
- ivala2081/vibe-game★ 4same name
- clnnn/playbook★ 0same name
- lunaman81/claude-skills★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository82 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- design/CLAUDE.mdCLAUDE.md
- docs/CLAUDE.mdCLAUDE.md
- src/CLAUDE.mdCLAUDE.md
- studio-tests/CLAUDE.mdCLAUDE.md
- adoptSkill
- architecture-decisionSkill
- architecture-reviewSkill
- art-bibleSkill
- asset-auditSkill
- asset-specSkill
- balance-checkSkill
- brainstormSkill
- bug-reportSkill
- bug-triageSkill
- changelogSkill
- code-reviewSkill
- consistency-checkSkill
- content-auditSkill
- create-architectureSkill
- create-control-manifestSkill
- create-epicsSkill
- create-storiesSkill
- day-one-patchSkill
- design-reviewSkill
- design-systemSkill
- dev-storySkill
- estimateSkill
- gate-checkSkill
- helpSkill
- hotfixSkill
- launch-checklistSkill
- localizeSkill
- map-systemsSkill
- milestone-reviewSkill
- next-stepSkill
- onboardSkill
- patch-notesSkill
- perf-profileSkill
- playtest-reportSkill
- project-settingsSkill
- project-stage-detectSkill
- propagate-design-changeSkill
- prototypeSkill
- qa-planSkill
- quick-designSkill
- regression-suiteSkill
- release-checklistSkill
- retrospectiveSkill
- reverse-documentSkill
- review-all-gddsSkill
- scope-checkSkill
- security-auditSkill
- settingsSkill
- setup-engineSkill
- skill-improveSkill
- skill-testSkill
- smoke-checkSkill
- soak-testSkill
- sprint-planSkill
- The first 60 shown.
The file
We split this skill into the rules it gives an agent
10 rules. 0 are common, 4 are distinctive, and 6 are very specific to this project.
0
Common
Most of the 5 edited versions we read keep these.
4
Distinctive
This author's own choices. Worth a look.
6
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 24
Ask with the question tool.
- Very specific: line 33
Run bash .claude/scripts/cut-read.sh and keep its rows:
- Distinctive: line 47
Read the brief itself too: its Question, Hooks, First minute, Challenge, Goal & fail state and Build order give the reasons.
- Very specific: line 58
Otherwise,: run the light check once, as .claude/skills/prototype/references/one-shot-4-change.md section 7 does before accept: its step 1 …
- Very specific: line 95
Target: ask in plain words named for the goal: when to release it (sell), when the Early Access launch is (early_access), when the pitch or …
- Very specific: line 98
Freeze: run bash .claude/scripts/cut-read.sh --target <date>.
- Distinctive: line 119
Leave out the Dates line for a goal that gets none, and Cut?
- Distinctive: line 120
Every file is named on the Writes line before writing, and nothing is written before the pick.
- Very specific: line 128
Run bash .claude/scripts/cut-apply.sh with --brief <path> for each brief row and:
- Very specific: line 143
Commit when git rev-parse --is-inside-work-tree succeeds, the pick being the commit instruction: git add <each wrote path, design/ ones …
“Common” is measured: kept by at least 6 in 10 of the 5 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
!bash "${CLAUDE_SKILL_DIR}/../../hooks/yaml-helper.sh" resolve_config --keys automation
Resolved above — use as-is. No block → defaults in .claude/docs/config-resolution.md.
Cut
Finishing a game is choosing what not to build. /cut reads what exists, proposes cuts with a
reason taken from it, and keeps a polish list; the person picks every cut. For sell,
early_access and publisher it asks a target date and proposes a freeze date. It writes the
brief, the polish list and any cut story's Status line, and makes one commit.
It runs only when asked: /cut, or "what should I cut?", "how do I finish this?". It works on
the game design/game-brief.md describes. A project with GDDs and no brief is /scope-check's.
Ask with the question tool. Without one, list the options numbered and end your turn to wait. The
question is the last thing in the reply. A skipped or dismissed question is not an answer: ask
again in plain text. Every ask follows .claude/docs/automation-modes.md: in collaborative and
guided the cuts and the dates are asked, as scope changes are major decisions; in autonomous
/cut writes the polish list only and leaves the proposals in its report: cuts and dates wait for
the person, and nothing is committed.
Phase 1: Read what exists
Run bash .claude/scripts/cut-read.sh and keep its rows:
| Rows | Use |
brief, title | The briefs written, both copies while they are the same text. brief none: say "/cut works from the brief's Features list, and there is no design/game-brief.md", name /scope-check when design/gdd/ has files, and stop. brief_differs: the game's folder copy differs; only design/game-brief.md is written, said in one line |
goal, dates | dates asked or not_asked follows the goal. goal - unset (or unknown): guess it with /start's goal table (.claude/skills/start/SKILL.md, Phase 4) from the brief's In your words, hobby when nothing fits, and show it guessed |
feature, keep, avoid, out, cut | The Features list by number, with its marks; the lists the proposals check |
recorded, passed | Dates already on the brief's Dates line, and any date already past |
game, record, open | A one-shot's last check record and its fails not marked deliberate |
smoke | A project's smoke command, from the engine table, or none with why |
feel | The feel log |
polish, polish_row | The polish list's rows and ticks. polish other_game: it is another game's; say so and write no polish rows |
story, stories_other | Stories traced to a Features item by number (cut: one already cut); a count of stories traced elsewhere, which /cut leaves alone (say so) |
Read the brief itself too: its Question, Hooks, First minute, Challenge, Goal & fail state and Build order give the reasons.
Fresh first: a
record ... stalerow means the game changed after its check record. A changed game gets the light check, never the small check again: the small check is the build's, andone-shot-check.sh --recordrecords only that, so never run it here. First see whether a light check is newer than every game file. For each marker on disk,evidence/changes.tsv(a change's light check) andevidence/cut-end.png(an earlier /cut's), runfind prototypes/<name> -type f -newer prototypes/<name>/evidence/<marker> ! -path '*/evidence/*' ! -path '*/dist/*' ! -name brief.md ! -name feel-log.md ! -name polish-list.md.- One prints nothing: that light check is fresh, so run nothing. Errors and a bot stop on
changes.tsv's last line that the polish list lacks become rows. - Otherwise, run the light check once, as
.claude/skills/prototype/references/one-shot-4-change.mdsection 7 does before accept: its step 1 (keep) and step 2 (play it once), the shotprototypes/<name>/evidence/cut-end.png. Anot_assessedrow: say "not checked:<why>" in one line. Fix nothing: each error, bot stop andchangedorgonekeep row becomes a polish row.
These rows end
· from: light check of <date>. The stalerecordrow stays; itsopenrows still count.- One prints nothing: that light check is fresh, so run nothing. Errors and a bot stop on
- A project:
smoke <command>: runtimeout 900 <command>once, as it is: no pipe, redirect or second command (the permission is for that command alone). A failure is a polish row (from: smoke run of <date>); a pass adds none.record nonewith no smoke command to run: say in one line that the polish rows come from the feel log and the person.
Phase 2: Propose
Cuts, in this order, each with a reason from what exists:
- A
not_yetitem that Out of scope or Avoid already rules out. - A
builtitem whose check fails (open), offered as cut it or polish it. - A
not_yetitem that no Question, Hook, First minute, Challenge, Goal & fail state or Keep line names.
For a selling goal, every not_yet item left also shows "build it before <freeze date>, or cut
it". Never propose an item a Keep line names. The person can still name any item.
Marks brought up to date: a not_yet item whose story row is Complete is shown built and
marked [x] with the other writes.
Polish rows come from what exists, never invented: each open row, each smoke failure, and what the person adds. Each row ends
· from: <source>: <check> check of <date>, smoke run of <date> or you said. The person's rows and ticks stay; a row whose
source now passes is ticked, said in one line. Rows about a cut item go with the cut.
Phase 3: Dates, for sell, early_access and publisher only
Any other goal gets no date, asked or shown.
- Recorded dates are shown, not asked again ("move the dates" changes them). A
passed freezerow is said in one line; apassed targetrow is asked again. - Target: ask in plain words named for the goal: when to release it (
sell), when the Early Access launch is (early_access), when the pitch or demo is (publisher). Turn a month or "next spring" into a date and show it. - Freeze: run
bash .claude/scripts/cut-read.sh --target <date>; itsproposed freezerow is 70% of the way from today to the target. Ask "Freeze on<date>?" Options:Freeze then/Another date. Say what it means: from then on no new features, only the polish list and fixes; recorded, and nothing holds you to it yet.target_past: ask the target again. - A goal changed on the screen ("I want to sell it") is written at once with
bash .claude/scripts/config-set.sh project.goal <value>, and its dates are asked. A guessed goal left as it is is not written. No otherproject.yamlkey is touched.
Phase 4: The screen and the pick
/cut · <title> · Goal: <goal> (set | guessed)
Cut? You pick; nothing is cut without your pick.
1. [ ] <item>
· <reason>
Staying: <n> built features · Never offered: your <n> Keep lines
Polish: <n> rows, from <the sources>
Dates: target <date> (<what>) · freeze <date>
Writes: <the briefs>, <the polish list>, <each story> · one commitLeave out the Dates line for a goal that gets none, and Cut? when there is nothing to propose. Every file is named on the Writes line before writing, and nothing is written before the pick. Then ask with the question tool, multi-select, up to four proposals ("Cut which? Pick none to cut nothing"); more are numbered in text and picked by number. The reason is the person's words if they gave any, else the proposal's. With dates to ask, ask the target in the same call, then the freeze as Phase 3 says.
Phase 5: Write and commit
Run bash .claude/scripts/cut-apply.sh with --brief <path> for each brief row and:
--cut <n> "<reason>"for each picked item: it leaves Features for the end of Out of scope as- [-] <item> · cut: <reason> (<date>), plus· was builtwhen it was[x]; its story's Status becomesBlocked — BLOCKED: cut by /cut <date>: <reason>, a status/dev-storystops on (a Complete story keeps its status:story moved).--mark <n>for each mark brought up to date.--dates <target> <freeze> "<what>"when dates were given.--polish "<row>",--tick "<row>"and--drop "<row>"for the polish list.
Moving an item renumbers the Features after it, so the stories traced to them are renumbered too
(story renumbered); the Writes line names every story a proposed cut can touch. bad rows mean
nothing was written: fix the arguments and run it again. A cut built item stays in the game until
it is taken out, an ordinary change: say "say 'take out <item>' to remove it".
Commit when git rev-parse --is-inside-work-tree succeeds, the pick being the commit instruction:
git add <each wrote path, design/ ones first>, then git commit -m "docs: cut <items> · dates ·
polish list (<n> rows)" -m "Brief: design/game-brief.md" -- <the same paths>. Leave out
cut <items> · when nothing was cut, and dates · when the Dates line did not change. No git:
"Not committed: this project has no git." Say in one line what was written.
Next step: build what is left before the freeze (/create-stories writes a story for each
not-yet item in a project), and work down the polish list.
Afterwards
- "Bring back
<item>":cut-apply.sh --back "<item>" built|not_yetmoves the line back to Features:builtwhile it is still in the game. Its story goes back to Ready. Commit thewrotepaths asdocs: bring back <item>. - "Move the dates": Phase 3 again, with the recorded dates shown, then write and commit as Phase 5 does.
- Nothing blocks or warns on the freeze date in this version: only /cut reads it back.
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
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.

