AGENT SKILL

core-data-and-geometry

Routes PhiFlow field, geometry, mesh, scene, and data-format workflows.

License 许可
MIT
最近更新
2026-09-03
Stars
★ 351
Forks
27

01 前言 Front Matter

name
core-data-and-geometry
license
MIT
metadata
{ "disco-role": "operating" }
description
Routes PhiFlow field, geometry, mesh, scene, and data-format workflows.
disable-model-invocation
true

02 资源文件 3

03 SKILL.md 正文

Core Data and Geometry

Use this sub-skill for the core PhiFlow objects that most other workflows build on: fields, grids, geometries, meshes, scenes, and field/scene I/O.

Route here for

  • CenteredGrid, StaggeredGrid, PointCloud, Noise, and AngularVelocity
  • Box, Sphere, Cuboid, Cylinder, UniformGrid, Mesh, Graph, Heightmap, SDFGrid, and SDF
  • Scene.create(), Scene.at(), Scene.list(), Scene.write(), Scene.read(), and scene cleanup helpers
  • field.read(), field.write(), field.resample(), field.sample(), and field.reduce_sample()
  • Geometry transformations and set operations such as shift, rotate, union, and intersection
  • scene / field file formats and legacy compatibility issues

Do not route here

  • installation or backend setup -> installation-and-backends
  • PDE stepping, fluids, waves, FLIP, or SPH -> physics-and-simulation
  • gradients, Jacobians, or inverse design -> optimization-and-learning
  • plotting, display, controls, or scalar logs -> visualization-and-ui

Start with these imports

from phi.flow import *
from phi import field, geom, math
from phiml.math import batch, channel, instance, spatial, vec

Most common decisions

  1. Grid type: use CenteredGrid for scalar quantities and StaggeredGrid for velocity-like vector fields.
  2. Sampling vs resampling: use field.sample() when you need values at a Geometry; use field.resample() or Field.at() when you need compatible field representations.
  3. Scene I/O: use Scene.create() for new scenes, Scene.write() to store fields, and Scene.read() to recover them. The scene format stores sim_ folders with *.npz field files.
  4. Geometry syntax: prefer keyword constructors such as Box(x=1, y=1) or the explicit slicing form Box['x,y', 0:1, 0:1].
  5. Mesh / implicit geometry: use Mesh for unstructured surfaces or volumes, and SDFGrid / SDF when you want implicit shapes.
  6. Legacy code: treat Domain as compatibility-only. New workflows should construct grids directly with dictionaries and extrapolation= or boundary= values.

Verified signatures to rely on

  • Scene.create(parent_directory, shape=(), name='sim', copy_calling_script=True, **dimensions)
  • Scene.write(self, data=dict=None, frame=0, **kw_data)
  • Scene.read(self, *names, frame=0, convert_to_backend=True)
  • field.read(file, convert_to_backend=True) / field.write(field, file)
  • field.resample(value, to, keep_boundary=False, **kwargs)
  • field.sample(field, geometry, at='center', boundary=None, **kwargs)
  • field.reduce_sample(field, geometry, **kwargs)
  • Box(*args, **kwargs) and Box['x,y', ...]
  • geom.Box, geom.Sphere, geom.Cylinder, geom.Mesh, geom.Graph, geom.Heightmap, geom.SDFGrid, and geom.SDF

Bundled references and helper

From this sub-skill directory, run the smoke helper when you want to verify the field / scene I/O path:

python scripts/scene_roundtrip.py --resolution 16

The helper writes temporary scene data, reads it back through both the scene API and field.read(), and cleans up afterwards.

04 同仓库更多 Skills

VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-triton-opt

Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-solve

END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-skill

PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-pattern-fusion

Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-orchestrate

Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-feedback

Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).

★ 351