How to use the Design Game Encounters agent skill

What the skill makes the agent do

Without a skill, an agent asked for a wave of enemies tends to pick a count and a spawn timer. This skill changes the starting point: it asks the agent to design encounters as decisions the player makes, not as lists of actors.

The source describes four areas of work. The agent composes an encounter by defining the objective, available space, enemy roles, spawn timing, hazards, player resources, exits, failure recovery and reward, adding one pressure source at a time. It checks fairness by protecting readable paths, telegraphs, camera sight lines and recovery windows, capping simultaneous committed attackers, and avoiding offscreen damage, unavoidable chains, or resets that duplicate rewards. It validates by creating deterministic starts for low resources, each wave, boss phase, victory and death/retry, and by testing desktop and mobile at the real camera distance. Finally it adjusts composition based on observed player decisions rather than raw completion time.

Install the skill from the ZIP

Download the ZIP for design-game-encounters from the page you are reading. The skill folder contains two files: SKILL.md, which holds the instructions, and agents/openai.yaml.

Option one is a personal install. Unzip the download into ~/.claude/skills/ so that the design-game-encounters folder sits inside it.

Option two is a project install. Unzip it into the .claude/skills/ folder of your project so it applies only to that codebase.

Option three is the Claude apps. Upload the ZIP under Settings → Capabilities → Skills.

• Personal: ~/.claude/skills/design-game-encounters/
• Project: .claude/skills/design-game-encounters/ inside your project
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Trigger the skill with natural requests

You do not need a special command. The skill is matched to requests about its topics: arena layout, enemy composition, spawn pacing, objectives, boss phases, reward cadence, encounter fixtures and difficulty validation. Name the game engine and the thing you want, and state whether you want design, implementation, tuning or testing.

Be specific about your situation. Mention the game's stack, the camera setup, and any resources the player tracks, such as health, ammo or dodge charges, so the agent can reason about them.

• "Design a three-wave arena encounter for my Three.js action game where the player has limited health pickups."
• "Review this boss fight's phase 2 for unavoidable damage and blocked sight lines."
• "Write deterministic test starts for low health, each wave, and the retry flow of this encounter."
• "Check whether this encounter gives duplicate rewards when the player dies and restarts."

What a good result looks like

Expect the agent to describe an encounter in terms of the player's choices: where they can move, which enemy forces a decision, when pressure arrives, and what the reward is. It should also flag fairness risks such as an attacker that can hit from offscreen or a retry that hands out the same reward again.

Treat the output as a design draft and a review checklist. The source is a short set of principles. It does not provide engine code, asset generators or a library of encounter templates, so the agent will still need to write any implementation itself and you will need to play-test it.

• A list of the encounter's objective, space, roles, pacing, resources, exits and reward
• Named fairness risks tied to the arena or attack pattern
• Deterministic test starts for each wave, boss phase, victory and death/retry
• Adjustments based on what players did, not only how long the fight took

Safety notes: scripts and network

According to the listing, this skill requires nothing beyond the agent itself, does not run scripts, and does not use the network. It consists of SKILL.md and agents/openai.yaml. Because it runs no code, installing it only adds instructions the agent can read.

The skill is released under the MIT licence. Still, read SKILL.md before installing any skill in a project you care about, since the agent will follow its instructions.

• Requires: nothing listed
• Runs scripts: no
• Uses network: no
• Licence: MIT

Tips for better encounter designs

Ask for one pressure source at a time when you are iterating. The skill itself recommends adding pressure in that way, so a staged request matches its method.

Ask for test setups alongside the design. The validation guidance is most useful when the agent produces named deterministic starts you can reproduce in your own build.

Test on real hardware at the real camera distance. The skill calls for desktop and mobile testing, and the agent cannot see your game's feel.

• Iterate one change at a time
• Request deterministic starts with every design
• Judge pacing from observed player decisions, not completion time