Agent Playtesting logo

Agent Playtesting

Organization
SummerEngine
agent-playtesting

Playtest by driving the live game: deterministic launch, frame-stamped probes, scripted or recorded input.

Overview

PublisherSummerEngine
Repositorysummer
Skill nameagent-playtesting
Stars
91
Forks
8
Bundled files
1
LicenseMIT
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • 1 bundled files

    Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

  • Open source

    Published by SummerEngine on GitHub. Read the source before you install it.

Installation

Install the Agent Playtesting AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/SummerEngine/summer/tree/main/library/skills/agent-playtesting
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/SummerEngine/summer.git /tmp/summer
mkdir -p .claude/skills
cp -r /tmp/summer/library/skills/agent-playtesting .claude/skills/agent-playtesting
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Agent Playtesting in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Agent Playtesting on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Agent Playtesting is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Agent Playtesting

Overview

playtesting-a-feature says a feature is not done until someone has played it. This skill is how you play it — not a hidden probe that dies when it finishes, but the live game, driven one action at a time, with a frame-stamped record of what each action did.

The engine's runtime-control ops make the running game a thing you can see and touch:

ToolDoes
summer_playStart the game. With instance + mode:"offscreen" a disposable parallel instance; with seed / fixed_fps (and deterministic:true offscreen) a reproducible one.
summer_game_probeState AND pixels of ONE frame, atomically: live tree, property reads, screenshot, all stamped with the same frame counters. The evidence tool.
summer_game_controlpause, resume, step exactly N physics/process frames, speed, instances.
summer_game_inputscript timed synthetic input; record_start / record_stop real input into res://.summer/replays/; replay a recording (with seed on a deterministic instance).
summer_runtime_set / summer_runtime_callSet a property / call a method on a live node. Actions, not evidence.
summer_runtime_spawnSpawn a PackedScene into, or free a node from, the live game.
summer_runtime_animateRead/drive an AnimationPlayer, an AnimationTree state machine, or read Skeleton3D bone poses.
summer_stopStop the main game, or instance to stop one offscreen instance.

Every one of these needs a RUNNING game and answers game_not_running otherwise. On an engine build that predates them the result is engine_lacks_op naming the fallback — usually the RunVerification probe from playtesting-a-feature. Record that as the expected outcome, not a failure.

The Loop

  Launch deterministically → Wait for boot → Probe BEFORE → Act → Step (or run) → Probe AFTER → Assert → Stop

1. Launch deterministically

Decide before pressing play what the run must pin.

  • Embedded main game (summer_play with no instance): the game the user sees. Add seed and fixed_fps to pin the RNG and the frame clock — seed only applies to a game that is not already running (determinism.applied: false with reason: already_running means summer_stop first).
  • Offscreen instance (summer_play {instance:"a", mode:"offscreen"}): a hidden child process the user never sees, addressed by name on every runtime tool. Up to three at once. deterministic: true launches it with --fixed-fps 60 --summer-seed <seed> --audio-driver Dummy; this is the only kind of instance on which summer_game_input action:"replay" accepts a seed.
  • Two variants side by side: instance:"before" and instance:"after" (different scenes or different summer_runtime_set setups), the same recorded input replayed into each, probes compared frame for frame.

The result's determinism.seed_scope tells you what the seed does NOT pin: RandomNumberGenerator instances (self-randomized on construction), scripts that call randomize(), wall-clock reads, thread timing. A feature that uses any of those is not deterministic under a seed; assert ranges there, or fix the script to use the global RNG.

2. Wait for boot — never sleep a guessed delay

Boot time varies. Confirm, do not assume:

  • Main game: summer_is_running until it reports running.
  • Offscreen instance: summer_game_control action:"instances" until the instance shows attached: true. Before that, every runtime op answers request_failed ("session has not attached yet"). session_timeout means the child never attached and was killed — read summer_get_console for its stderr.

3. Probe BEFORE

summer_game_probe
  tree: {path: "/root/Main", depth: 2}
  props: ["/root/Main/Player:position", "/root/Main/Player:velocity", "/root/Main/HUD/Health:value"]

Write down frame.process_frames and frame.physics_frames. Every path you will act on comes from this tree — runtime paths are absolute (/root/Main/Player) and differ from the edited scene when nodes spawn, rename, or reparent at runtime. Look at the screenshot and say what is in it.

4. Act

One action per step, so the AFTER probe attributes cleanly:

  • summer_runtime_set path:"/root/Main/Player" property:"position" value:"Vector3(0, 2, 0)" — read applied. applied: false (not_applied) means a script rewrote the value the same frame, or the literal type was wrong; do not proceed as if it landed.
  • summer_runtime_call path:"/root/Main/Boss" method:"take_damage" args:[25] — return is what the method answered; the effect still needs the probe.
  • summer_game_input action:"script" events:[{at_frame:0, type:"action", action:"move_right", hold_ms:500}, {at_frame:30, type:"action", action:"jump", hold_ms:50}] — rejected[] names events the game refused (unknown_action = not in the InputMap; bind it with summer_input_map_bind and restart, or drive type:"key"). One script in flight per instance: busy means wait for the first.
  • summer_runtime_spawn action:"spawn" parent:"/root/Main/Enemies" scene:"res://enemies/goblin.tscn" props:{position:"Vector3(4, 0, -2)"} — use the returned node.path from here on.

5. Step — for exact assertions

When "after 3 physics frames the velocity is zero" is the claim, do not let wall-clock time decide how many frames ran:

summer_game_control action:"pause"
summer_game_probe props:[...]                 # frame N
summer_runtime_set ... / summer_game_input ... wait:false
summer_game_control action:"step" frames:3   # kind "physics" (default)
summer_game_probe props:[...]                 # frame N+3 — the step draws its last frame before replying

The step result carries before, after, exact, and overshoot; cite them. The game stays suspended after a step — summer_game_control action:"resume" when you want it to run again. Time-based behaviour (a 500 ms hold, a 2 s cooldown) runs on the game's clock: with fixed_fps a step of N frames IS N/fps seconds; without it, use at_ms and read clock_mapping ("approximate" means the mapping went through the physics tick rate).

A minimized game window draws no frames and cannot step (timeout). game_breaked means the game sits at a script breakpoint; continue it in the debugger first.

6. Probe AFTER, then assert

Same probe as step 3. Compare the two:

  • positions, velocities, HUD values — as inequalities (x > before.x + 50), never exact equality unless the run is deterministic AND frame-stepped;
  • tree — the spawned node exists under its parent; the freed one is gone (a queue_free takes effect at the end of the frame, so step once before asserting absence);
  • the screenshots — look at both and describe the difference in words. A tree that says "goblin spawned" and a frame that shows no goblin is a bug, not a pass.

Then summer_get_debugger_errors — a feature that "worked" while logging null-reference errors did not work.

7. Stop

summer_stop (main) or summer_stop instance:"a" for each offscreen instance you started. Offscreen children also die when the editor exits, but do not leave them for the user to discover.

Input scripts vs. recordings

action:"script"record_start / record_stop / replay
SourceYou write it from the specWhatever actually happened — a human's hands or a script's output
ReadableYes; edit a frame number and re-runA JSON file of events; treat it as an artifact
Timingat_frame exact; at_ms exact only with fixed_fpsRecorded frames re-based on the first event
Best forGolden paths, edge probes, regression checks"Do exactly what the user did", A/B of two builds under identical input
ReproducibleWith seed + fixed_fps on the instanceWith seed on a deterministic:true offscreen instance (replay refuses seed elsewhere: nondeterministic_instance)

Recording caps at 20,000 events / about 1 MiB (truncated: true when hit). Files land in res://.summer/replays/<id>.json; pass that exact path to replay. A waited script or replay is capped at 20 s by the engine — for longer sequences pass wait:false and follow along with probes.

What seed does not pin

Read determinism.seed_scope on the play result and believe it. Pinned: the global RNG (randi, randf, randi_range, randf_range, randfn, Array.shuffle, Array.pick_random), re-seeded after SceneTree init and before autoloads and the main scene. Not pinned: RandomNumberGenerator instances, scripts calling randomize(), wall-clock reads (Time.get_ticks_msec used as a source of variation), thread timing, and anything read from the network or the file system. When a replay diverges on a deterministic instance, one of these is the reason — find it in the script before doubting the tooling.

Honesty rules

  • Cite frames. "At frame 1834 the player was at x=12.0; after move_right for 30 frames, at frame 1864 x=57.3" is a verification. "The player moved" is not.
  • applied: false is a no. Do not narrate a value change the read-back denied.
  • completed: false is a no. The script did not finish inside the wait; probe to see how far it got, do not call it done.
  • rejected[] is part of the result. Report the rejected events and why, even when the rest applied.
  • engine_lacks_op is the expected outcome on an engine without these ops. Say so, fall back to the playtesting-a-feature probe loop, and do not describe a runtime playtest you could not run.
  • A screenshot you did not look at is not a screenshot. Describe what is in it.
  • Parallel instances are not the user's game. Findings from instance:"a" describe that instance's launch (scene, seed, fixed_fps); say which instance a claim came from.
  • Do not hand the walkthrough to the user while any of these tools works. Ask a human only for what a probe cannot judge — feel, beauty, fun (debugging-game-feel).

Red flags — STOP

Red flagReality
Asserting motion from a summer_runtime_set resultapplied means the write landed; probe to see the motion
One probe, then "it works"One frame proves a state, not a change; two frames prove a change
Sleeping a fixed delay after summer_playBoot varies; summer_is_running / instances attached:true
position.x == 250.0 on a non-deterministic runAssert inequality, or pin the run with fixed_fps + steps
Ignoring rejected[] because applied > 0The rejected event may be the one the feature needed
seed on the embedded game called "deterministic"It pins the global RNG only; deterministic:true needs an offscreen instance
Leaving offscreen instances runningStop each one you started

Related skills

  • playtesting-a-feature — the iron law (no shipped claim without a played walkthrough) and the RunVerification probe loop this skill replaces when the runtime tools exist.
  • verifying-scenes — evidence discipline for the EDITED scene (snapshot, diff, screenshot).
  • verification-before-completion — the code layer that must be green before any of this.
  • debugging-game-feel — when the feature works but feels wrong.
  • investigating-bugs — root-causing a reproducible bug once a recording reproduces it.

From a shell

No Summer MCP tools in this session? Run each one as a command; when summer --version is missing or below 4.0, use npx -y summer-engine@latest in place of summer. Fields are flags in kebab-case (--scene-path res://main.tscn) or one --input '<json>', and summer ops describe <op> prints the exact input. The summer-cli skill has the rest.

In this skillCommand
summer_game_controlsummer engine game-control
summer_game_inputsummer engine game-input
summer_game_probesummer engine game-probe
summer_get_consolesummer engine get-console
summer_get_debugger_errorssummer engine get-debugger-errors
summer_input_map_bindsummer engine input-map-bind
summer_is_runningsummer engine is-running
summer_playsummer engine play
summer_runtime_animatesummer engine runtime-animate
summer_runtime_callsummer engine runtime-call
summer_runtime_setsummer engine runtime-set
summer_runtime_spawnsummer engine runtime-spawn
summer_stopsummer engine stop

Bundled files

The model reads these on demand while the skill is loaded. They are exposed as readable files and are never executed.

Frequently asked questions

What does the Agent Playtesting AI skill do?

Playtest by driving the live game: deterministic launch, frame-stamped probes, scripted or recorded input.

Why use Agent Playtesting on TypingMind?

Because you install it once and use it with any model. Agent Playtesting is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Agent Playtesting in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/SummerEngine/summer/tree/main/library/skills/agent-playtesting. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Agent Playtesting?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Agent Playtesting?

As many as you like. As long as a model supports skills, you can use Agent Playtesting with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Agent Playtesting AI skill free?

Yes. It is published on GitHub by SummerEngine under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇