agentleFS
kotaroyamazakikotaroyamazaki/godot-agent-kit/skills/playtest/SKILL.md

playtest

Skill2 starsChanged 13 days agoMITView on GitHub

Find out where a first-time player gets lost in a Godot game, by having a subagent that knows nothing about the game play it from screenshots and report what it pressed and why. Use for "is this understandable on first play", "does the tutorial work", "playtest it", "are the controls confusing", and after building or changing UI, onboarding or a tutorial. 「初見で分かるか見て」「遊び方が伝わるか確かめて」「プレイテストして」などで使う。

5 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions5 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 5

    1. linggen/linggen★ 11same name
    2. sebnado/worldbuild-bench★ 9same name
    3. WhorideChicken/first-playable★ 3same name
    4. callmeradical/skills★ 0same name
    5. StephaneBayle/coach-rugby★ 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 repository7 other files it gives its agents.

The file

We split this skill into the rules it gives an agent

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

  • 0

    Common

    Most of the 4 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    After fixing, run it again with a new subagent.

    line 25
  2. Distinctive:

    "Nothing reacted" is a finding, not an agent mistake: If the agent says a press did nothing but the game did change state (a counter went …

    line 62

“Common” is measured: kept by at least 6 in 10 of the 4 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.

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

Blind playtest

Whoever built the game knows the answers, so they cannot judge whether it explains itself. Hand it to a subagent that has been told nothing about the game.

Steps

  1. Fix the entry point: the launch arguments (a fresh save, if the game has such a flag) and the viewport size from project.godot. Check that godot-kit shot works in the game. Give the agent a clean copy. The folder name, CLAUDE.md, README.md and docs/ all tell it what the game is. Copy the project somewhere with a neutral name (/tmp/playtest/g1) without those files, and point the brief there.
  2. Spawn exactly one subagent (a small, cheap model is enough). Give it only the brief below. Do not give it the game's name, genre, rules, README or source, and tell it not to read them.
  3. Compare its report with what you intended. Candidates to fix: things it never noticed were pressable, things it misread, places it stopped.
  4. After fixing, run it again with a new subagent. The previous one is no longer a first-time player.
  5. Keep every round. Save the report as docs/playtests/NN.md with the date and what you changed afterwards, so you can see whether round 2 lost fewer places than round 1. examples/sleepy-cat/playtests/ has two real rounds.

Brief (use as is; fill in the <...> parts and delete the input line that does not apply)

You are playing a game for the first time. There is no manual. Do not read source code, README files or docs.
Your job: look at the screen, press what looks pressable, and work out what this game is.

Tool (run it inside <absolute path of the game folder>):
  godot-kit shot <out>.png --after <sec> [--every 1 --count N] -- <launch args> <inputs...>
- The screen is <width>x<height>. Input coordinates use that size; times are seconds since launch.
- Saved images can be smaller than the screen (the window is scaled down). Convert image coordinates to input
  coordinates by multiplying by <width> / <image width>.
- Touch game:           --tap=x,y,sec   --hold=x,y,sec,length   --swipe=x1,y1,x2,y2,sec
- Mouse/keyboard game:  --click=x,y,sec  --move=x,y,sec (hover)  --key=Name,sec  (Enter, Space, Escape, Left, Right, Up, Down, A...)
- Every launch starts from the beginning. So each time, repeat ALL inputs you have made so far and add one new input
  at the end (one launch replays the whole session).
- Read the saved images to see the result.

Procedure:
1. Launch with no input and capture 2 s and 6 s after start. Write what you see, what looks pressable, and what you
   think the game is about.
2. Choose one thing, act on it, capture 1 s later. Write what you expected and what actually happened.
3. Repeat for up to 12 actions. If you do not know what to do, say so, and say what you pressed at random.

Report (facts only; no praise, no suggestions):
- Per action: the input / why you chose it / expected / actual
- Things you thought were pressable that did not react
- Things you only later realized were pressable
- The goal and rules of the game in your own words (or "I could not tell")
- The moment you were most lost

What to look at in the report

  • "Nothing reacted" is a finding, not an agent mistake. If the agent says a press did nothing but the game did change state (a counter went down, a timer reset), the change was invisible: a feedback bug. Confirm by replaying the same input yourself and comparing with a run without it (cmp a.png b.png works when nothing on screen animates). Idle motion makes it worse: the agent cannot tell a small reaction from the animation.
  • Did it hesitate before the first action? Then pressable things do not look pressable before interaction.
  • Could it state the goal and the rules? If not, the game is missing a place that teaches them in words.
  • Actions where "expected" and "actual" differ usually lack motion, sound or a clear beat.

Limits

  • It sees still frames, so it cannot judge sound or how good movement feels. Ask the person to play on a real device for that.
  • Games that need fast or precisely timed input are better checked with a fixed script of inputs (the check skill).

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.