Skip to content

CLAUDE.md

Latest commit

 

History

History
55 lines (43 loc) · 3.02 KB

CLAUDE.md

File metadata and controls

55 lines (43 loc) · 3.02 KB

Agent Instructions

User Context

  • Name: James Cotton
  • Role: Physician-Scientist
  • Domains: Biomechanics, Rehabilitation, Motion Capture

Guidelines

You MUST consult and follow the guidelines in skills/enforce-guidelines/references/. The /enforce-guidelines skill activates automatically before code tasks.

The single source of truth for tool configuration (ruff, pytest, coverage) is templates/pyproject.toml. Use /audit-repo to check if a repo has drifted from it.

Workflow

  1. During work: Follow guidelines in skills/enforce-guidelines/references/
  2. Before commits: Run /deslop to clean AI artifacts
  3. After completion: Verify with ruff format . && ruff check . && pytest

Available Skills

Mandatory (Auto-Activate)

  • /enforce-guidelines — Ensures all work follows RAE guidelines. Activates automatically.

Utility

  • /deslop — Clean AI-generated slop from staged changes
  • /consult-guidelines — Review relevant guidelines for current task
  • /scaffold-repo — Initialize a new repository with correct structure
  • /golden-test-harness — Self-generating golden/parity tests: config-driven harness regenerates goldens; tests/golden/ is gitignored, not committed
  • /audit-repo — Check current repo against RAE standards, report drift, fix gaps
  • /config-improvement — Propose improvements to upstream RAE repo
  • /bead-driven-development — Orchestrate planning + execution with beads tracking (optional, disabled by default)
  • /investigation — Scaffold structured research in scratch/
  • /excalidraw — Generate Excalidraw diagrams, render to PNG/SVG, visually verify
  • /semi-formal-code-reasoning — Structured reasoning templates for patch equivalence, fault localization, code QA, and code review
  • /github-activity-review — Periodic cross-repo GitHub activity review with thematic organization

Domain-Specific

  • /datajoint-biomechanics-schema — DataJoint pipeline schema reference
  • /pose-datajoint — Python code patterns for DataJoint pose queries
  • /fetching-synchronized-data — Temporal alignment of keypoints with kinematic reconstruction
  • /gait-metrics — Gait analysis: walking segments, spatiotemporal metrics, GDI
  • /jax-config — JAX/Equinox-compatible config setup with pytree registration and tyro CLI
  • /jax-memory-and-retracing — Diagnose/avoid GPU & host OOMs, memory leaks, and JIT retracing in JAX/Equinox
  • /camera-model — Camera projections, intrinsics/extrinsics, triangulation, mm/m unit conventions
  • /efficient-rendering — Fast, correct rendering of meshes/point clouds/overlays: EGL headless GPU, OpenCV↔GL conventions, distortion strategies, chunked JAX rasterizers, video compositing
  • /gait-lab-dataset — Clinical gait lab schema, video-mocap sync, MMC-GaitLab trial matching

CRITICAL: NEVER Modify Database Entries

See skills/enforce-guidelines/references/coding-standards.md for the full rule. In short: DO NOT call update1(), delete(), drop() on any DataJoint table. Database entries are shared lab state.