Index of every page on this site: llms.txt
All skills
Run it yourself

Builds a throwaway playable prototype in about an hour to answer one question, usually "is this mechanic fun?": a one-question brief, a greybox, a debug HUD, a playtest and a keep, kill or retune call. Use for "prototype this", "just get something playable" or whether to keep prototype code. Not for a timed jam (`game-jam`) or level blockouts (`level-design`).

Use this Skill: https://skilld.dev/gh/gamedev-skills/awesome-gamedev-agent-skills/prototype-fast

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈95 tokens for metadata: the name and description. ≈1.4k when used: this file.

Description uses 4.7% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈379 characters, or 4.7%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

Prototype Fast

Get a mechanic playable fast to learn whether it is worth building. The output is a decision (keep, kill or retune) written down in the repo, not a product. The agent writes the brief, builds the smallest thing that answers it, instruments it, and asks the user to play; the user's report decides.

When to use

  • Validating one mechanic or "what if", a quick slice of one interaction, or de-risking an idea before committing real time.

Not for: a jam with a deadline and submission (game-jam); level layout and blockout (level-design); shipping (steam-publish, itch-publish); the production build of a validated feature (engine or genre skill).

Workflow

  1. Write the brief first to prototypes/<idea>/BRIEF.md (Pattern 1) and confirm it with the user: one question, the core verb, keep and kill signals, a timebox. Two questions means two prototypes; one that tests everything decides nothing.
  2. Default to throwaway. Hard-code values, one file is fine, no architecture. Choose "keep" structure only when the user says the code will grow, because prototypes that pretend to be production code rot into the shipping codebase.
  3. Contain it. Build in prototypes/<idea>/ (a separate project or folder the main game never imports) or on a throwaway branch.
  4. Greybox everything but the verb. Boxes, circles, capsules, the engine's default font, no audio or one debug beep. Spend nothing on what the question does not depend on.
  5. Instrument the question. Draw a debug HUD with the values being judged (speed, jump height, timing window, time to first success) and expose the tuning values as named constants at the top of the file, so a retune is a one-line change.
  6. Run it yourself first (Verify), then ask the user to play without explanation and report: did they do the core verb unprompted, how long it took, did they replay, what felt off. If someone else can play, ask the user to watch them and report the same.
  7. Make the call against the brief and append it to BRIEF.md: keep (schedule a rewrite in the real project), kill (one line on what was learned), or retune (change one constant and play again).

Patterns

1. BRIEF.md

QUESTION:   The one thing this must prove, ideally yes/no.
CORE VERB:  The single action the player repeats.
THROWAWAY:  yes (default) | no, because ...
TIMEBOX:    e.g. 60 min; stop when it ends, even if unfinished.
KEEP IF:    observable signal (see Pattern 3).
KILL IF:    observable signal.
RESULT:     <date> keep | kill | retune - evidence from the playtest

2. Only the verb is real (engine-neutral pseudocode)

const RUN_SPEED = 8.0, JUMP_VELOCITY = 12.0, GRAB_WINDOW = 0.15   # tune here

on update(dt):
    read_input()
    apply_core_mechanic(dt)                  # the only thing being tested
    draw_box(player)                         # a box, not a sprite
    draw_debug_hud(speed, airtime, window_hits, time_to_first_success)
    # enemies, menus, saves, audio, art: stubbed or absent

3. Keep, kill or retune

KEEP when, with placeholder art:
  - a fresh player does the core verb on purpose within about 30 s, unprompted, and
  - they replay the loop without being asked.
KILL when:
  - the verb only feels good after it is explained, or
  - making it fun needs systems far beyond the prototype, or
  - art or levels are being added to avoid admitting the verb is flat.
RETUNE when one value is clearly off (too slow, window too tight): change it, play again.

Verify

Run these yourself before handing the prototype to the user:

  1. It launches from the command line (the engine's run or headless command, or a local web server on 127.0.0.1) with no errors in the log or console.
  2. The loop completes: start, perform the verb, succeed or fail, restart. Drive it with a scripted input or a short automated run where the engine allows; otherwise play it once.
  3. The debug HUD shows the values named in the brief and they change as expected.
  4. git status (or a search for imports) shows nothing outside prototypes/<idea>/ changed and the main game does not import it.
  5. BRIEF.md has a RESULT line after the user's playtest.

Gotchas

  • Polishing too early. Art, menus and audio make a flat verb look finished and delay the verdict.
  • No kill criteria. Without a written "kill if", every prototype "has potential" and nothing is cut. Decide the signal before anyone is attached to the code.
  • Building systems instead of the mechanic. Inventory, saves and settings are not the question. Stub them.
  • The spike becomes the product. A keep authorizes a rewrite in the real project, not a copy-paste of prototype code and its assumptions.
  • Prototyping inside the real codebase. It tangles experiments with shipping systems and makes the spike expensive to delete.
  • Only the author playtests. The author knows the controls and the intent. One uncoached player reveals more in two minutes than an hour of self-play, so always ask the user for a report rather than judging it yourself.

Related skills

  • game-jam: the same scope discipline under a deadline with a submission.
  • level-design: when the question is about a level's layout rather than a mechanic.
  • game-feel: juice, only after the verb is proven.
  • Engine cores (godot-gdscript, phaser-core, love2d-core, unity-csharp-scripting, ...) for the greybox itself and for the rewrite after a keep.

Source: SKILL.md on GitHub

No rule matched.

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No alerts8d3 checks · Risk SAFE
  • Gen Agent Trust Hub9d

    The skill provides a workflow for rapid game prototyping. It is considered low risk as it involves generating and running local scripts and processing user playtest reports, which creates a surface for indirect prompt injection.

  • Socket8d

    No alerts

  • Snyk9d

    Risk: LOW · No issues

Signed by skilld at c6ca6c4. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub last week.

Activeupdated last week

README badge

README badge for gamedev-skills/awesome-gamedev-agent-skills/prototype-fast