
Game combat skill polish
Scores and improves game combat skills and spells using blind judging of filmed footage
- Repository
- MengTo/Skills
- Stars
- 7,187
- Licence
- MIT
- Files
- 13 files
What it does
Teaches an AI agent a structured workflow for reviewing and improving the visual and timing quality of video game attacks and spells. The agent films casts from a game in a headless browser on a stepped clock, builds contact sheets and close-ups, has two fresh judge subagents score them blind against a readability or art rubric, and then recommends or applies fixes such as making results land on the hit frame or modelling a hero prop. It is aimed at game projects, not general coding.
Great for
- Game developers polishing spell, ability or attack animations and effects
- Technical artists who want repeatable, blind-scored reviews of combat visuals
- Teams rebuilding a skill to match a named reference game
Skip it if
- Non-game projects, since it assumes a game dev server with review-only hooks that must be built for each game (the skill cannot work without them)
- Quick one-off visual feedback, because the full workflow filming, judging and rounds takes many steps
Try asking
- “Score all the combat skills in my game out of 10 for readability and art”
- “Rebuild the fire nova spell so it reads as itself and the damage lands on the impact frame”
- “Make my warrior's charge attack look more like the reference, and keep iterating until it scores 7 or higher”
Install in 3 steps
- 1
Download the skill
The ZIP contains the 13 original files, the MIT licence and an attribution note.
game-dev-combat-skill-polish.zip - 2
Add it to your agent
Claude Code (all projects)
mkdir -p ~/.claude/skills && unzip game-dev-combat-skill-polish.zip -d ~/.claude/skills/Claude Code / Agent SDK (this project only)
mkdir -p .claude/skills && unzip game-dev-combat-skill-polish.zip -d .claude/skills/Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.
- 3
Ask for the task
No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.
Get better results
- Create a project reference file (references/<project>.md) with your review hooks, harness settings and scores so far, so the agent does not have to rediscover them.
- Always name which rubric a score comes from, and treat the lower of two blind judge scores as the result.
- Read the scripts and set the environment variables (BASE, OUT, HOOK) yourself before running film.mjs, and point it only at a local dev server.
What's inside
- references/rubric-art.md
- references/rubric-readability.md
- scripts/beforeafter.py
- scripts/close.py
- scripts/film.mjs
- scripts/judge-set.py
- scripts/merge.py
- scripts/perf.mjs
- scripts/probe.mjs
- scripts/scenarios.example.mjs
- scripts/strips.py
- SKILL.md
- templates/judge-prompt.md
Redistributed unmodified from MengTo/Skills under the MIT licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.