- Generated CC0-style prototype grid textures (0.5 m cells) and a shared
LevelMaterials factory: world-space triplanar mapping over every code-built
level (test level, dust2, procedural arena) with the existing color coding
kept as tints. Readable surfaces at speed instead of flat color boxes;
materials are cached per tint so identical surfaces batch.
- .gdignore + .gitignore the raw downloaded asset packs in addons/ whose
overlong paths broke Godot import scans and spammed git warnings.
- Replaced the rotted FSM test file (invalid call(t) on Callables, stateless
stub nodes, hung forever without quitting) with a real suite: 10 tests
covering state registration, jump bookkeeping, wall-run momentum
preservation and upward-carry cap, per-player dash cooldown (regression for
the old shared static), dash speed stacking, chain soft cap, landing
events, and slide entry momentum. New headless entrypoint:
godot --headless --path . -s res://movement/tests/run_fsm_tests.gd
Co-Authored-By: Claude Fable 5 <[email protected]>
- Quake-style air-strafe accelerate (real speed gain from strafing; external
speed from dash/rockets never clamped)
- Wall run: keeps entry momentum (incl. upward carry), accelerates along the
wall instead of hard-setting velocity, gravity fades in over the run
- Slide: slope physics (gravity projected along floor accelerates downhill),
flat-ground decel instead of multiplicative friction, entry boost, direct
slide->wallrun and slide->dash transitions
- Dash: cooldown 10s -> 2s, per-player (was a static shared across instances),
momentum fully kept on exit, FOV punch event
- Grapple: taut pendulum with active rope control (W reels in, S pays out),
release keeps swing energy
- Wall climb: carries upward momentum in, jump-off kick, shared vault helper
- Crouch capsule: single owner in the machine, smoothly lerped, ceiling check;
camera eye height follows via crouch factor (no more head snapping)
- Landing: machine emits 'land' events scaled by impact; camera dip + pitched
down thud; footsteps get pitch variation
- Camera rig: event-driven (land dip, dash FOV kick, vault pitch impulse),
slide roll tilt
- Model: Jump vs Fall split by vertical velocity, Land one-shot on heavy
landings, wall-run body lean (synced to remotes via synced_wall_side)
Co-Authored-By: Claude Fable 5 <[email protected]>
Layer a procedural SkeletonModifier3D (ShooterPoseModifier) on top of the base
clip so the character reads like a movement-shooter avatar:
- Directional lean: banks into the movement direction (right/left/back) and
blends smoothly for diagonals, driven by velocity relative to facing.
- Slide: leans the torso back and pitches the head up to look forward, instead
of the base clip's forward-torso/legs-out "spine break".
- Airborne: plays a composed Jump pose (weapon ready) rather than a flailing
fall.
- Weapon hold: the base clip already keeps the arms down (weapon at the hip);
on ADS both arms lift and swing in toward centre-front to aim. ADS is read
from the active weapon and synced (synced_is_ads) so remote players raise
their weapons too.
The controller feeds movement direction + ADS to SkinnedPlayerModel.set_locomotion()
each frame (works for local and remote via synced velocity/rotation). All
rotations are authored in skeleton space (fwd=+Z, up=+Y, right=-X) and converted
per-bone; tuning constants are at the top of ShooterPoseModifier.
Verified by rendering the poses (idle/strafe/back/slide/ads) in a real Godot
viewport. Smoke test 30/30.
Co-Authored-By: Claude Fable 5 <[email protected]>
Character pipeline (tools/):
- sketchfab_import.py: search/download via Sketchfab Download API with
license/attribution tracking (SKETCHFAB_API_TOKEN)
- autorig.py: headless Blender auto-rig — fits a Mixamo-named skeleton
to unrigged humanoids, binds automatic weights
- merge_animations.py: merges the shared Mixamo-skeleton animation
library onto any rigged character, strips root motion, canonical names
- pipeline.py: one command chains download -> rig -> animate -> register
In-game skin system:
- SkinManager autoload reads skins.json (auto-written by the pipeline)
- SkinnedPlayerModel rewritten: canonical clips with fallback chains,
blend times, speed-scaled locomotion, weapon bone attachment
- First-person: full animated body for the owner, head hidden via
SkeletonModifier3D; third-person: full model for other players
- Skin selector in main menu; skin id synced in multiplayer
- Fixed GLBLoader crash (GLTFDocument.get_animation_count doesn't exist)
Multiplayer sync overhaul:
- Movement is now client-authoritative: the owning peer simulates
locally (no input round-trip), server keeps health/kills/death
- Remote players interpolate synced_position/velocity with
extrapolation and snap-on-teleport
- Knockback/impulses routed to the simulating peer
Audio:
- AudioManager autoload: SFX/Weapons/Footsteps/UI/Music buses, pooled
3D players, variation + pitch randomization, auto-registration from
assets/sounds
Docs: 3D_ASSET_PIPELINE.md rewritten end-to-end, new ASSET_SOURCES.md
(non-procedural animation/map/sound sources) and SOUND_DESIGN.md.
Verified with debug/spawn_smoke_test.gd (headless: 24/24 checks pass).
Co-Authored-By: Claude Fable 5 <[email protected]>
The Miku model T-posed because the previous _update_skin() used
set_bone_pose(i, get_bone_pose(i)) which is a no-op when the pose
is unchanged. Replace with skeleton.notification(NOTIFICATION_UPDATE_SKELETON)
to force the Skeleton3D to recompute bone transforms and notify the
MeshInstance to update vertex buffers.
Also fix scale: the GLB is ~1.2m tall (not 1.6m as documented), so
set scale_factor=1.25 to fit the 1.8m player capsule, and add
position_y_offset=-0.2 to align the model's feet with the player origin.
Co-Authored-By: Hermes Agent <[email protected]>
The skinned player model was loading miku_rigged_final.glb which only
contained a single "Idle" animation (1 track, near-rest pose). This caused
"animation 'Jump/Walk/Run' not found" spam and a T-pose appearance.
Switch to miku_rigged_animated.glb which has all 7 animations
(Crouch, Death, Idle, Idle2, Jump, Run, Walk).
Additional fixes:
- Fix scaling: the GLB is already in meters (~1.6m tall), not cm.
Removed the unconditional 0.01 cm->m conversion that was shrinking
the character to 1/100th size.
- Fix bone track paths: use "MikuRig/Skeleton3D:" prefix matching the
actual scene tree structure (AnimationPlayer is sibling of MikuRig).
- Add _ensure_locomotion_animations() to create Walk/Run/Jump/Idle
procedurally if the GLB is missing them.
- Eliminate per-frame "not found" log spam with rate-limited warnings.
- Update level_runtime.gd and test_level_builder.gd to use animated GLB.
Closes #3d-player-model animation pipeline
The Blender GLTF exporter was stripping animation tracks (only 1 of 51
channels survived export). Now SkinnedPlayerModel detects this at
runtime and creates a proper Idle animation programmatically using Godot's
Animation API with 5 bone tracks (Spine, Arms, Neck, Hips).
Also switched model_path back to miku_rigged_final.glb since the
re-export wasn't adding usable animations.
- Remove duplicate is_local variable in level_runtime.gd line 224
- Remove orphaned animation code from old load_model in skinned_player_model.gd
- Both files now compile cleanly
Godot headless/CLI can't import GLB files — the GLTF loader only works
in the editor GUI. Switched SkinnedPlayerModel to use GLTFDocument.append_from_buffer()
which parses GLB data from raw bytes at runtime.
Verified: GLB parses successfully, generates scene with MikuRig skeleton,
AnimationPlayer, and 7 animations (Crouch, Death, Idle, Idle_001, Jump, Run, Walk).
Also added characters/glb_loader.gd as a reusable utility for loading
any GLB file at runtime.
- Add miku_rigged_animated.glb (3.27MB) with 18-bone skeleton and 6 animations
(Idle, Walk, Run, Jump, Crouch, Death) via Blender auto-rig script
- Add tools/rig_and_animate.py: reusable Blender script for rigging any humanoid
mesh with Mixamo-compatible bone naming
- Update SkinnedPlayerModel: scale_factor, first-person mode, animation state
matching (Idle/Walk/Run/Jump/Crouch/Death)
- Update level_runtime.gd: use skinned model for local player, procedural
humanoid for remote players
- Update skin_manager.gd and test_level_builder.gd to use animated model
- Fix Godot 4.2.1 'is not Type' syntax in 5 weapon files
- Add editor/import_miku_to_tscn.gd for editor-based GLB import
- Fix tab character in _build_elevated_walkways() call (line 63)
- Fix untyped 'pos' variable in grapple pillar loop (line 376)
- Remove fog_sky_affinity (Godot 4.3+ only property)
- Replace preload() with load() for runtime script loading
- Replace 'is DoubleBarrelShotgun' type check with duck-typing
- Fix 'is not OfflineMultiplayerPeer' syntax in grenade spawn
- Replace absf() with abs() in state_machine, state_air, state_ground
- Update test_level.tscn to remove stale UID reference
All fixes verified: PARSE OK and scene runs without script errors
on Godot 4.2.1 headless.
- state_wall_run.gd: typed ray_origin/ray_end as Vector3
- test_level_builder.gd: restructure builder to put MovementStateMachine
+ state children under Player, attach camera under Player too
- player_movement_controller.gd: @onready uses get_node so type is not inferred
- test_level.tscn: minimal Node3D root (zero SubResource references)
- debug/test_level_builder.gd: runtime floor/walls/lighting/camera/UI
shapes/meshes created from BoxShape3D/BoxMesh — no .tscn subresources
- project.godot: revert features to 4.6 (installed build is 4.6.3)
- Removed all SubResource BoxShape3D references from test_level.tscn
- Changed PackedStringArray feature from 4.6 to 4.2
Scene is now pure nodes + transforms, no external resource IDs
- scenes/test_level/test_level.tscn — playable test environment (30x30 floor + 4 walls + ramp)
Floor at y=-1, Player spawn at (0, 1.5, 8), FpsCamera at head height (y=1.7)
DirectionalLight + WorldEnvironment + UI label overlay (speed/state/chain)
- debug/debug_test_env.sh — launch wrapper for manual playtesting
No code changes: raw physics testbed. Intended to be opened as first scene in Godot window.
Closes#2 (test-level scene task)
Hermes