MelonS/MelonS-Agents

An agent that builds, plays, and verifies its own game ? a colony-sim prototype gated by 15-scenario input-level repro tests + isolated-grader rubric verdicts ? plus production media skills: music-video (60s 9:16 shorts, beat-aligned, genre-aware) and Korean job-board digest. agentskills.io-spec compliant. v0.4.0.

16Stars on the repository
38Mods indexed here, across every type
10d agoLast push, which is what freshness is scored on
MITLicence, which decides whether bodies are shown

auditor

01

MelonS/MelonS-Agents

Agent Claude Code

Repository-wide read-only auditor. Periodically (or on demand) reviews the whole project for drift — docs vs code, roadmap freshness, operator-contract compliance, cost-model accuracy, stale TODOs, secret leakage. Writes a structured report to docs/audit/ .md. Does NOT modify code; only diagnoses and proposes fixes.…

16 10d ago A 82 tokens original MIT

content-director

02

MelonS/MelonS-Agents

Agent Claude Code

Orchestrates the content-shorts pipeline 리서치팀 → 제작팀 ⇄ 법률팀 → 출시팀 for a single short (info/news/idol). Delegates each stage to its team subagent, runs the legal feedback loop until PASS or max iterations, and reports the final mission folder. Invoke from /info-short, /news-short, /idol-short.

16 10d ago A 79 tokens original MIT

cut-judge

03

MelonS/MelonS-Agents

Agent Claude Code

A video-cut reviewer that opens generated frames and scores them against a supplied scene brief, captions, and constraints.

16 10d ago A 86 tokens original MIT

editor

04

MelonS/MelonS-Agents

Agent Claude Code

Applies transformations and produces final mission outputs. Reads plan.md + resources/MANIFEST.md, writes deliverables to outputs/. Invoke after resourcer.

16 10d ago A 33 tokens original MIT

game-ai-designer

05

MelonS/MelonS-Agents

Agent Claude Code

AI Designer (specialist). Decision trees, utility AI, behavior selection, NPC routines. Activated for colony sims, RTS, tactical RPGs — any genre where NPCs make independent choices. Not for simple "approach + attack" enemies (that's game-designer territory).

16 10d ago A 63 tokens original MIT

game-artist

06

MelonS/MelonS-Agents

Agent Claude Code

Visual Artist. Produces sprites, UI visuals, animation states. Picks between procedural gen, Kenney CC0 fetch, or SDXL last-resort. Owns visual consistency with Director's tone. Triggered after Designer's sprite list is set.

16 10d ago A 54 tokens original MIT

game-build-engineer

07

MelonS/MelonS-Agents

Agent Claude Code

Build Engineer / Technical Artist. Owns the Unity batchmode pipeline — SceneSetup, BuildScript, AutoScreenshotter, asset import discipline. Triggered after Programmer + Artist have outputs ready, before QA can verify.

16 10d ago A 48 tokens original MIT

game-designer

08

MelonS/MelonS-Agents

Agent Claude Code

Game Designer. Decomposes the Director's vision into concrete game systems, mechanics, controls, balance. Outputs the scripts/systems list that ends up in the genre YAML. Triggered after Director and before Programmer.

16 10d ago A 47 tokens original MIT

game-director

09

MelonS/MelonS-Agents

Agent Claude Code

Game Director. Owns the game's vision, tone, and core feel — "what kind of game is this?" Not the same as PM (PM owns schedule). Triggered when starting a new genre, when team subagents conflict on direction, or when a build's "feel" diverges from intent.

16 10d ago A 66 tokens original MIT

game-level-designer

10

MelonS/MelonS-Agents

Agent Claude Code

Level Designer (specialist). Spawn patterns, wave composition, boss patterns, biome layout. Activated for horde survivors, tower defense, twin-stick shooters, action. Not for grid puzzles / casual.

16 10d ago A 48 tokens original MIT

MelonS/MelonS-Agents

Agent Claude Code

Narrative Designer (specialist). Story beats, NPC dialogue, event flavor text, lore. Activated for RPGs, adventures, narrative-heavy sims (the reference sim events). Not for arcade / casual. Korean + English bilingual capable.

16 10d ago A 58 tokens original MIT

game-pm

12

MelonS/MelonS-Agents

Agent Claude Code

Game Producer / PM. Owns schedule, milestones, dependencies, scope. NOT the same as Game Director (Director owns vision/tone). Triggered at start of every prototype (allocate days), at end of each Day (advance milestone), or when scope changes.

16 10d ago A 57 tokens original MIT

game-programmer

13

MelonS/MelonS-Agents

Agent Claude Code

Gameplay Programmer. Translates Designer's script list into actual C# files using the template catalog. Knows every bug pattern baked into templates and refuses to write code that would re-trigger them. Triggered after Designer and before Build Engineer.

16 10d ago A 51 tokens original MIT

game-qa

14

MelonS/MelonS-Agents

Agent Claude Code

Game QA. Runs the autonomous self-verify loop (build → launch .exe → screenshot → read PNG → verdict). Catches regressions across Days. Triggered after every Day's Build Engineer pass.

16 10d ago A 45 tokens original MIT

game-sound-designer

15

MelonS/MelonS-Agents

Agent Claude Code

Sound Designer. Produces BGM + SFX. Procedural-first (gen-sfx CLI), Suno/external when game-feel demands it. Owns the throttle discipline that prevents per-frame buzz. Triggered when Designer's audio list is set.

16 10d ago A 59 tokens original MIT

MelonS/MelonS-Agents

Agent Claude Code

Systems Designer (specialist). Focused on resource economies, needs decay, progression curves, balance numbers. Activated only for genres with non-trivial systems depth — colony sims (the reference sim), city builders, deep RPGs, factory games. NOT activated for casual puzzle / arcade prototypes.

16 10d ago A 67 tokens original MIT

legal-team

17

MelonS/MelonS-Agents

Agent Claude Code

법률팀 — reviews a rendered content short before release. Runs the deterministic license gate (copyright.sh guardpublish via legal-gate.sh) AND content-legal judgment (factual accuracy, defamation, unverifiable claims, synthetic-narration disclosure, trademark/IP; for idol/real-subject shorts also…

16 10d ago A 108 tokens original MIT

note-judge

18

MelonS/MelonS-Agents

Agent Claude Code

A review agent that scores a creative video plan before image or video production begins. It checks the opening hook, structure, emotional progression, audience participation, visual feasibility, and legal or disclosure concerns.

16 10d ago A 98 tokens original MIT

orchestrator

19

MelonS/MelonS-Agents

Agent Claude Code

Main orchestrator. Receives missions, delegates to planner/resourcer/editor/qa subagents in sequence, aggregates results, and writes the mission summary to $RECORDSDIR. Use for any multi-step mission that spans planning → resourcing → editing → validation.

16 10d ago A 58 tokens original MIT

planner

20

MelonS/MelonS-Agents

Agent Claude Code

Mission planner. Breaks a mission brief into concrete steps, identifies required resources/tools, defines acceptance criteria for QA. Outputs plan.md. Invoke first in any mission flow.

16 10d ago A 36 tokens original MIT

production-team

21

MelonS/MelonS-Agents

Agent Claude Code

제작팀 — renders the 60s 9:16 content short by driving agents/missions/content-short/run.sh (faceless core + profile + idol subject overlay). Consumes research.json; on a legal REVISE verdict, applies the fix list and re-renders. Second stage of the content-shorts pipeline; loops with the 법률팀.

16 10d ago A 76 tokens original MIT

qa

22

MelonS/MelonS-Agents

Agent Claude Code

Validates mission outputs against plan.md acceptance criteria. Runs regression checks. Writes qa-report.md with pass/fail per criterion. Invoke last in mission flow.

16 10d ago A 33 tokens original MIT

release-team

23

MelonS/MelonS-Agents

Agent Claude Code

출시팀 — packages a legally-cleared content short for manual upload. Runs only after a legal PASS. Produces per-platform title/description/tags, a thumbnail, the attribution manifest, the disclosure lines, and a publish checklist. Never auto-uploads (operator uploads manually; public URLs stay out of the repo). Final…

16 10d ago A 77 tokens original MIT

research-team

24

MelonS/MelonS-Agents

Agent Claude Code

리서치팀 — gathers verified facts + license-screened media hints for a content short (info/news/idol). Produces resources/research.json (claims tied to fact-sources, recency stamp, B-roll visual terms, risk flags). First stage of the content-shorts pipeline. Separates fair-use fact citations from license-gated media…

16 10d ago A 93 tokens original MIT