agentleFS
EL4CTEOEL4CTEO/roblox-skills/skills/roblox-npc-ai/SKILL.md

roblox-npc-ai

Skill92 starsChanged 48 days agoMITView on GitHub

NPCs and enemy AI - PathfindingService (CreatePath, ComputeAsync, costs, modifiers, links, blocked paths), chase/follow loops, state machines, perception and line of sight, spawning/pooling, scaling to many NPCs. Use when creating enemies, pets, followers, shopkeepers, wave spawners, or fixing NPCs that get stuck, jitter, or lag the server.

2 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

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

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Use Humanoid for NPCs that need walking/jumping/climbing.

    line 16
  2. Very specific:

    Add reaction delays and field-of-view checks (root.CFrame.LookVector:Dot(direction.Unit) > 0.5) so NPCs feel fair.

    line 168

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “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

NPCs and AI

Architecture

  • NPC logic runs on the server (authoritative targets, damage, drops). Visual extras (idle animations, nameplates, effects) can run on clients.
  • One NPC service drives all NPCs from a single loop (e.g. 5–10 Hz "think" rate, staggered), not one script + while true per NPC. Tag NPC models (CollectionService) and keep per-NPC state in a table.
  • Server-owned physics: npc.PrimaryPart:SetNetworkOwner(nil) after spawning, so NPCs don't stutter when ownership moves between nearby players (and exploiters can't push them).
  • Use Humanoid for NPCs that need walking/jumping/climbing; for simple movers (turrets, floating enemies, crowds) use an AnimationController + Animator and move with constraints or CFrames.

Pathfinding

--!strict
local PathfindingService = game:GetService("PathfindingService")

local path = PathfindingService:CreatePath({
	AgentRadius = 2,
	AgentHeight = 5,
	AgentCanJump = true,
	AgentCanClimb = false,
	WaypointSpacing = 4,
	Costs = {
		Water = 20, -- avoid water (Enum.Material name)
		DangerZone = math.huge, -- PathfindingModifier.Label: never cross
	},
})

-- Walk `humanoid` to `goal`. Returns true on arrival, false if unreachable or interrupted.
local function walkTo(humanoid: Humanoid, goal: Vector3): boolean
	local root = humanoid.RootPart
	if not root then
		return false
	end
	local ok = pcall(function()
		path:ComputeAsync(root.Position, goal)
	end)
	if not ok or path.Status ~= Enum.PathStatus.Success then
		return false -- NoPath: try a fallback (direct MoveTo, new goal, teleport if stuck long)
	end

	local waypoints = path:GetWaypoints()
	local blocked = false
	local blockedConnection = path.Blocked:Connect(function(blockedIndex)
		blocked = true -- an obstacle appeared on a waypoint ahead; caller recomputes
		print("path blocked at waypoint", blockedIndex)
	end)

	for index = 2, #waypoints do
		local waypoint = waypoints[index]
		if waypoint.Action == Enum.PathWaypointAction.Jump then
			humanoid.Jump = true
		end
		humanoid:MoveTo(waypoint.Position)
		local reached = humanoid.MoveToFinished:Wait() -- false after 8 s timeout
		if not reached or blocked or humanoid.Health <= 0 then
			blockedConnection:Disconnect()
			return false
		end
	end
	blockedConnection:Disconnect()
	return true
end

return walkTo
  • CreatePath + ComputeAsync is the current API (FindPathAsync is superseded). Reuse Path objects per agent type; ComputeAsync yields — never call it every frame for every NPC.
  • Chasing moving targets: recompute when the target moved more than a few studs since the last compute or every ~0.5–1 s, and skip pathfinding entirely when there's a clear line of sight (workspace:Raycast from NPC to target) — just MoveTo the target.
  • PathfindingModifier (child of a part/region, Label matching a Costs key, PassThrough for doors) and PathfindingLink (between two attachments, e.g. ladders, teleporters, boats; the waypoint has Label and Action = Custom) shape paths.
  • Limits: start→goal straight-line distance ≤ 3,000 studs, 20,000-node budget; far goals need intermediate targets. Visualize the navmesh with Studio's Navigation Mesh visualization option.
  • On clients, pathfinding sees only streamed-in geometry: compute paths on the server.

Behavior: state machine

--!strict
type State = "Idle" | "Chase" | "Attack" | "Return"

type Npc = {
	model: Model,
	humanoid: Humanoid,
	home: Vector3,
	state: State,
	target: Model?,
	lastAttack: number,
}

local AGGRO_RANGE = 40
local ATTACK_RANGE = 5
local LEASH_RANGE = 80
local ATTACK_COOLDOWN = 1.2

local losParams = RaycastParams.new()
losParams.FilterType = Enum.RaycastFilterType.Exclude

local function hasLineOfSight(from: Vector3, target: BasePart, ignore: { Instance }): boolean
	losParams.FilterDescendantsInstances = ignore
	local result = workspace:Raycast(from, target.Position - from, losParams)
	return result == nil or result.Instance:IsDescendantOf(target.Parent :: Instance)
end

local function think(npc: Npc, findTarget: (Vector3, number) -> Model?)
	local root = npc.humanoid.RootPart
	if not root or npc.humanoid.Health <= 0 then
		return
	end
	local targetRoot = npc.target and npc.target:FindFirstChild("HumanoidRootPart") :: BasePart?
	local distance = if targetRoot then (targetRoot.Position - root.Position).Magnitude else math.huge

	if npc.state == "Idle" then
		npc.target = findTarget(root.Position, AGGRO_RANGE)
		if npc.target then
			npc.state = "Chase"
		end
	elseif npc.state == "Chase" then
		if not targetRoot or (root.Position - npc.home).Magnitude > LEASH_RANGE then
			npc.state = "Return"
		elseif distance <= ATTACK_RANGE then
			npc.state = "Attack"
		elseif hasLineOfSight(root.Position, targetRoot, { npc.model }) then
			npc.humanoid:MoveTo(targetRoot.Position)
		end -- else: request a path (throttled) via walkTo in a separate thread
	elseif npc.state == "Attack" then
		if not targetRoot or distance > ATTACK_RANGE * 1.5 then
			npc.state = "Chase"
		elseif os.clock() - npc.lastAttack >= ATTACK_COOLDOWN then
			npc.lastAttack = os.clock()
			local targetHumanoid = npc.target and npc.target:FindFirstChildOfClass("Humanoid")
			if targetHumanoid then
				targetHumanoid:TakeDamage(10)
			end
		end
	elseif npc.state == "Return" then
		npc.target = nil
		npc.humanoid:MoveTo(npc.home)
		if (root.Position - npc.home).Magnitude < 4 then
			npc.state = "Idle"
		end
	end
end

return think

For complex AI (many behaviors, priorities), use a behavior tree or utility AI; keep the same "think at a fixed rate, act through small primitives" structure. See references/scaling-npcs.md for spawning, pooling, and performance with hundreds of NPCs.

Perception tips

  • Find targets with workspace:GetPartBoundsInRadius against a collision group/filter containing only characters, or iterate Players:GetPlayers() (cheaper for small counts) and compare squared distances.
  • Add reaction delays and field-of-view checks (root.CFrame.LookVector:Dot(direction.Unit) > 0.5) so NPCs feel fair.
  • Stagger thinking: npcIndex % 5 == frame % 5 spreads work across frames.

Related skills

roblox-characters-animation (humanoids, animation), roblox-physics (ownership, raycasts), roblox-performance (profiling, parallel Luau), roblox-game-systems (waves, combat, drops).

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.