game-ai Skill for Claude: NPC Decisions, Steering and Pathfinding
Makes Claude design NPC and enemy AI with state machines, behavior trees, steering and A* pathfinding, in any engine.
- Installs
- 9.3K
- on skills.sh
- Stars
- 6.4K
- on GitHub
- Files
- 2
- besides SKILL.md
- Updated
- Oct 2026
- last repo commit
This skill makes Claude build NPC behavior in three separate layers: decide what to do, steer how to move, and path how to route around the map. It helps you pick a finite state machine, a behavior tree or utility scoring, then shows the patterns: a state class, a behavior tree tick, seek and arrive steering, and an A* heuristic that stays admissible.
It is engine-neutral, with examples in Godot 4.7 GDScript and Python. For a specific engine's navigation API it points to other skills in the same collection. It bundles 2 reference notes and no scripts. It writes game code in your project, so review the changes.
Good for
- Game developers building enemy or NPC behavior
- Anyone choosing between a state machine, a behavior tree and utility scoring
- Godot, Unity and Unreal developers who want engine-neutral AI patterns
Install game-ai
Claude Code
- 1
Copy this prompt.
Install the game-ai skill in Claude Code. Run this command: claude plugin marketplace add gamedev-skills/awesome-gamedev-agent-skills claude plugin install gamedev@awesome-gamedev-agent-skills Then check that it loaded and tell me how to use it.
- 2
Paste it into Claude Code. It shows the change and asks before running it.
Pick one route. The gamedev plugin installs a router skill plus 77 skills. The disciplines plugin, listed in the repo marketplace.json, holds 15 cross-engine skills including this one. The skills CLI needs Node.js 22.20 or later, per the README.
From the project README. Source.
What's in the skill
When it loads
Design NPC and enemy decision-making with finite state machines, behavior trees, steering behaviors, and A* pathfinding — engine-neutral algorithms that pair with the detected engine's navigation API. Use when building enemy AI, an FSM or behavior tree, steering/flocking, or pathfinding, or when the user mentions state machine, behavior tree, blackboard, A*, navmesh, seek, or patrol/chase.
Sections in SKILL.md
- When to use
- Core workflow
- Patterns
- 1. Finite state machine (one state object, explicit transitions)
- 2. Behavior tree tick (composite nodes return a status)
- 3. Steering: seek and arrive (smooth, frame-rate independent)
- 4. A heuristic must not overestimate (or paths stop being shortest)
- Pitfalls
- References
- Related skills
Key facts
- Installs on skills.sh
- 9,272 (Nov 2, 2026)
- GitHub stars
- 6,427 (Nov 2, 2026, whole repo)
- Last repo commit
- Oct 28, 2026
- What installs
- 3 files including SKILL.md: references/pathfinding.md and references/behavior-trees.md, no scripts
- skills.sh audits
- Gen Agent Trust Hub pass, Socket pass, Snyk pass (Nov 2, 2026)
What it includes
Six steps from choosing a decision model to verifying by watching the agent
Two to five states with clear transitions suit an FSM. Many reactive, prioritized behaviors suit a behavior tree. Continuous scoring suits utility scoring.
Core workflowOne state object per behavior with enter, update and exit, and explicit transitions
Transition logic stays inside states, not in a growing pile of if flags. The example is in GDScript.
FSM patternSequence and selector nodes return SUCCESS, FAILURE or RUNNING
A guard example reads: chase if the player is visible, otherwise patrol. references/behavior-trees.md covers leaves, decorators and a blackboard.
Behavior tree tickForces, not teleports; arrive slows inside a radius so the agent stops cleanly
Scale by delta time every frame so movement stays frame-rate independent.
Steering: seek and arriveKeep the heuristic admissible: Manhattan for 4-way grids, octile for 8-way
references/pathfinding.md has the full A* loop, grid versus waypoint graphs and when to use an engine navmesh.
A* and its heuristicSix traps, from per-frame pathfinding to agents stuck on walls, each with a fix
Each has a short fix, such as recomputing paths on a timer and adding a stuck timeout.
Pitfalls
From the official docs
Permissions
Low risk
It guides Claude to write AI code in your game project. Review the diff before you commit.
What it can do
- Read your data
- Create and change things
- Delete things
- Read-only mode
Watch out for
It writes game code in your project
The skill bundles no scripts, but its job is to make Claude implement state machines, behavior trees and pathfinding. Check the changes like any other code.
Questions
What does the game-ai skill do?
It makes Claude design NPC and enemy AI in layers: decisions with state machines or behavior trees, steering for smooth movement, and A* pathfinding. It is engine-neutral.
How do I install it in Claude Code?
Use the install box at the top. The README documents the gamedev plugin for everything, and the box also lists the smaller disciplines plugin and a one-skill route.
Does it work in Godot, Unity and Unreal?
Its examples are in Godot 4.7 GDScript and Python, and it says it is engine-neutral. For concrete Unity and Unreal navigation APIs it points to the unity-navmesh and unreal-behavior-trees skills, and for Godot to NavigationAgent2D/3D.
Do I need the other skills in the collection?
No. SKILL.md names the related skills as the place to bind these patterns to a specific engine's API, movement feel or level generation. The algorithms and two reference notes are in this folder.
Does it work in the Claude app?
The install guide documents a ZIP upload for Claude.ai on paid plans. Upload the whole game-ai folder so its two reference notes come with it.
Is it safe?
The folder bundles no scripts. Claude will write AI code in your project, so review the diff like any other change.
Information
updated Nov 2, 2026
- Folder
skills/disciplines/game-ai
Sources
Docs, issue trackers and discussions this page is built from.
Show all 3Hide
- SKILL.md · github.com
- INSTALLATION.md · github.com
- README · github.com