Imported from thedivergentai/game-dev-agentic-skills (
skills/bevy-ai-navigation/SKILL.md). Install upstream withnpx skills add thedivergentai/game-dev-agentic-skills --skill bevy-ai-navigation. Copyright stays with the author.
Bevy AI navigation
When to use
Load this when an actor must chase, patrol, or wait. This skill owns the intent and the desired velocity. Paths stay on bevy-navigation. Transform integration stays on bevy-character-movement. Line-of-sight sensors stay on bevy-spatial-queries.
ECS shape
- Components: a mode flag can be an enum
Component(see bevy-state-machines for the pattern). Goal position can be your own component. No agent, radius, or layer type was opened. - Resources: a shared patrol table can be a
Resourceyou define. No navigation resource was opened. - Systems: read the mode and write a desired velocity or waypoint on
UpdateorFixed. Use.chain()so policy runs before movement. Do not call an agent callback that writesTransform. - Schedule:
Fixed/FixedMainwhen the result feeds the movement step.
Policy map
| Need | Owner |
|---|---|
| Chase vs patrol vs wait | Enum Component plus a system that writes Goal |
| Retarget throttle | Timer resource ticked on FixedUpdate |
| Crowd and avoidance | bevy-navigation |
| Bake and obstacles | bevy-navigation |
| Line of sight | bevy-spatial-queries |
Policy writes DesiredVelocity. It does not move Transform and it does not query a navmesh.
[dependencies]
bevy = "0.20"
#[derive(Clone, Copy, PartialEq, Eq, Component)]
enum Brain {
Wait,
Patrol,
Chase,
}
#[derive(Component)]
struct WaypointIndex(usize);
#[derive(Component)]
struct Goal(Vec3);
#[derive(Component)]
struct DesiredVelocity(Vec3);
#[derive(Component)]
struct Player;
#[derive(Resource)]
struct PatrolRoute {
points: Vec<Vec3>,
}
#[derive(Resource)]
struct RetargetClock(Timer);
fn tick_policy(
time: Res<Time>,
mut clock: ResMut<RetargetClock>,
route: Res<PatrolRoute>,
players: Query<&Transform, With<Player>>,
mut actors: Query<(&Brain, &Transform, &mut WaypointIndex, &mut Goal)>,
) {
clock.0.tick(time.delta());
if !clock.0.just_finished() {
return;
}
let player_pos = players.iter().next().map(|transform| transform.translation);
for (brain, transform, mut cursor, mut goal) in &mut actors {
match *brain {
Brain::Wait => goal.0 = transform.translation,
Brain::Patrol => {
if route.points.is_empty() {
continue;
}
cursor.0 = (cursor.0 + 1) % route.points.len();
goal.0 = route.points[cursor.0];
}
Brain::Chase => {
if let Some(pos) = player_pos {
goal.0 = pos;
}
}
}
}
}
fn steer(mut actors: Query<(&Transform, &Goal, &mut DesiredVelocity), With<Brain>>) {
for (transform, goal, mut velocity) in &mut actors {
let offset = goal.0 - transform.translation;
let flat = Vec3::new(offset.x, 0.0, offset.z);
velocity.0 = if flat.length_squared() < 0.05 {
Vec3::ZERO
} else {
flat.normalize() * 3.0
};
}
}
pub struct NavPolicyPlugin;
impl Plugin for NavPolicyPlugin {
fn build(&self, app: &mut App) {
app.insert_resource(PatrolRoute {
points: vec![Vec3::new(-6.0, 0.0, 0.0), Vec3::new(6.0, 0.0, 4.0)],
})
.insert_resource(RetargetClock(Timer::from_seconds(
0.4,
TimerMode::Repeating,
)))
.add_systems(FixedUpdate, (tick_policy, steer).chain());
}
}
#[derive(Component)]
struct PatrolAgent;
fn spawn_patrol(mut commands: Commands) {
commands.spawn((
PatrolAgent,
Transform::from_xyz(0.0, 0.0, 0.0),
DesiredDir(Vec3::X),
));
}
#[derive(Component)]
struct DesiredDir(Vec3);
fn hold_when_close(mut agents: Query<(&Transform, &mut DesiredDir), With<PatrolAgent>>) {
for (transform, mut desired) in &mut agents {
if transform.translation.distance(Vec3::ZERO) < 0.5 {
desired.0 = Vec3::ZERO;
}
}
}
Expert Knowledge Delta
Target: Bevy 0.20+.
Core ECS types: Component, Resource, Query, Commands, Startup, Update, Fixed / FixedMain (bevy::time), .chain(), Plugin, DefaultPlugins, MinimalPlugins. Messages and observers are an ownership slot. See bevy-events-messaging for their APIs. Systems run in parallel unless .chain() orders the ones that conflict.
Who owns state
| Piece | Owner | Who may mutate | Schedule |
|---|---|---|---|
| Data on one entity | Component |
One system via Query |
Update or FixedMain |
| One value for the whole app | Resource |
One system | Startup to insert, then the schedule that reads it |
| Spawn | Commands |
The spawning system | Startup for the initial world; later schedules for runtime spawn |
| Order that must not race | .chain() |
The chained systems only | Same schedule |
| Plugin boundary | Plugin |
build registers systems and resources |
Before the app runs |
Borrow rule: two systems that both mutably query the same component must be ordered (.chain()) or split so only one writes. Unrelated systems stay parallel. That is the ECS page rule, not a guess at a system-set API.
Decision
| If you need | Do this | Do not |
|---|---|---|
| Per-entity data | Component + Query |
One script that owns the whole actor |
| App-wide config or clock | Resource |
Copy the value onto every entity |
| Stable simulation step | FixedMain and the Fixed clock |
Tie simulation to raw frame Update |
| Startup spawn | Commands in Startup |
Spawn from a render system |
| Cross-system notice that is not shared state | Message or observer slot in bevy-events-messaging |
A new event type invented in this file |
| An API outside the documented set | Check the official docs first | A made-up crate or API |
Load references/ecs-ownership.md for the ownership table.
NEVER
-
Do not pin
/0.20.0/in docs URLs. WHY: links stay onlatest; the 0.20 target is stated once in prose. -
Do not mutably query the same component from two unordered systems. WHY: the schedule runs systems in parallel unless
.chain()orders them. -
Do not invent a physics, navigation, networking, or XR crate here. WHY: this skill only names documented Bevy crates and modules.
-
No navigation-agent or nav-server types from another engine. WHY: this skill writes
GoalandDesiredVelocity; a foreign agent type would moveTransformand skip the chain into movement. -
No named third-party navigation or physics crate. WHY: Bevy code stays idiomatic ECS and uses only documented Bevy APIs.
-
No invented path, bake, or avoidance APIs. WHY: Bevy code stays idiomatic ECS and uses only documented Bevy APIs.
Reference
The library pin on bevy-master stays 0.20.0. No AI navigation module was opened. Load references/ecs-ownership.md when splitting goal writes from velocity writes. Do not load it to bake a mesh.
Official Documentation
- Assets - Official asset catalog. No AI plugin was selected.
- Examples - Example tree on the
latestbranch. - ECS -
Component,Resourcepairing via bevy-data-resources, systems, and.chain(). - App states are not per-entity patrol modes. Per-entity flags stay components (bevy-state-machines).
Related Skills
Prerequisites
- bevy-master - ECS ownership.
- bevy-navigation - Paths and navigation meshes.
- bevy-state-machines - Enum components for a mode flag.
Complements
- bevy-character-movement - Integrates a waypoint or velocity.
- bevy-spatial-queries - Sensors and spatial queries.
Downstream / consumers
- None yet.
Master
- bevy-master - Bevy router.
