Browse skills

Merge multiple fine-tuned models using mergekit to combine capabilities without retraining. Use when creating specialized models by blending domain-specific expertise (math + coding + chat), improving performance beyond single models, or experimenting rapidly with model variants. Covers SLERP, TIES-Merging, DARE, Task Arithmetic, linear merging, and production deployment strategies.

Reduce LLM size and accelerate inference using pruning techniques like Wanda and SparseGPT. Use when compressing models without retraining, achieving 50% sparsity with minimal accuracy loss, or enabling faster inference on hardware accelerators. Covers unstructured pruning, structured pruning, N:M sparsity, magnitude pruning, and one-shot methods.

Train Mixture of Experts (MoE) models using DeepSpeed or HuggingFace. Use when training large-scale models with limited compute (5× cost reduction vs dense models), implementing sparse architectures like Mixtral 8x7B or DeepSeek-V3, or scaling model capacity without proportional compute increase. Covers MoE architectures, routing mechanisms, load balancing, expert parallelism, and inference optimization.

Accelerate LLM inference using speculative decoding, Medusa multiple heads, and lookahead decoding techniques. Use when optimizing inference speed (1.5-3.6× speedup), reducing latency for real-time applications, or deploying models with limited compute. Covers draft models, tree-based attention, Jacobi iteration, parallel token generation, and production deployment strategies.

Applies cognitive science frameworks for creative thinking to CS and AI research ideation. Use when seeking genuinely novel research directions by leveraging combinatorial creativity, analogical reasoning, constraint manipulation, and other empirically grounded creative strategies.

NIH grant research skill for clinical researchers. Grill-me intake (research idea + career stage + preliminary data + environment + submission posture + known institute targets) locks down the funding strategy before any search runs. Runs a 5-facet Consensus positioning analysis (with draft Significance/Innovation language), maps the research to the right NIH institutes and study sections via RePORTER, finds NOSIs and funded overlap, and produces an editable Word document (.docx) with budget/scope-aware mechanism recommendations, submission timelines, and a mandatory program officer recommendation. Use when the user asks about research funding or makes any grant-related request (e.g., 'grants for [topic]', 'find grants for my research idea', 'what grants match my research', 'help me find NIH funding', 'grant opportunities for my research'). NIH-only scope — non-NIH funders (PCORI, DOD CDMRP, VA, foundations) are out of scope and flagged at intake.

Use when reviewing, validating, or auditing Blender Python code for correctness. Runs systematic checks for deprecated API usage, context errors, version compatibility issues, threading violations, data reference invalidation, incorrect operator calls, and addon structure compliance. Prevents shipping code with silent version-dependent failures. Keywords: code review, validation, audit, deprecated API, context error, version compatibility, threading, addon structure, code quality, Blender Python, my addon has errors, check my code, why does my script fail.

Use when migrating, porting, or upgrading Blender Python scripts and addons across major versions (3.x to 4.x to 5.x). Provides a systematic migration process covering API renames, removed functions, changed parameters, extension system migration, BGL to GPU module conversion, and bone collection migration. Prevents incomplete migrations that compile but fail at runtime. Keywords: migration, porting, upgrade, version, 3.x to 4.x, 4.x to 5.x, API rename, bgl to gpu, extension migration, bl_info to manifest, bone collection, script stopped working after update, upgrade addon to new Blender.

Use when writing Blender Python scripts that access bpy module structure, RNA data, the context system, dependency graph, or operator invocation. Prevents the common mistake of accessing context attributes outside their valid scope or calling operators without checking poll(). Covers bpy.data, bpy.context, bpy.ops, depsgraph evaluation, and RNA property access patterns across Blender 3.x/4.x/5.x. Keywords: bpy, Blender Python, RNA, context, depsgraph, operator, bpy.data, bpy.ops, how to start scripting Blender, Blender Python basics, access objects in scene, get selected object.

Use when drawing custom overlays, viewport visualizations, or offscreen renders in Blender Python. Prevents the critical mistake of using the deprecated bgl module (removed in Blender 5.0) instead of the gpu module. Covers gpu.shader, gpu.batch, gpu.state, SpaceView3D draw handlers, built-in shaders, and BGL-to-gpu migration. Keywords: gpu module, bgl, draw handler, viewport overlay, offscreen rendering, SpaceView3D, shader, UNIFORM_COLOR, Blender Python drawing, draw on viewport, custom overlay, draw lines in 3D view.

Use when dealing with Blender Python runtime behavior -- mathutils, threading, handlers, timers, or crashes after undo/redo. Prevents the #1 runtime crash: using threading for bpy operations instead of bpy.app.timers, or caching bpy.data references across undo boundaries. Covers Vector/Matrix/Quaternion, KDTree, BVHTree, @persistent handlers, bpy.msgbus subscriptions, and background mode limitations. Keywords: mathutils, threading, undo, ReferenceError, bpy.app.timers, @persistent, bpy.msgbus, KDTree, BVHTree, background mode, Blender crash, startup script, auto-run, run on open, Blender freezes.

Datadog CLI for searching logs, querying metrics, tracing requests, and managing dashboards. Use this when debugging production issues or working with Datadog observability.

Use when implementing end-to-end BIM workflows that combine IfcOpenShell, Bonsai, and Blender -- such as IFC creation from scratch, model enrichment, validation pipelines, geometry extraction, or batch processing of building models. Prevents the common mistake of skipping unit and context setup before creating geometry, or directly modifying IFC attributes instead of using ifcopenshell.api.run(). Covers property set management across tools, spatial hierarchy patterns, and version compatibility for IFC2X3/IFC4/IFC4X3. Keywords: BIM workflow, IFC creation, model validation, property extraction, batch processing, spatial hierarchy, cross-technology, IfcOpenShell Bonsai Blender, pset management, how to create IFC model, step by step BIM, get properties from IFC.

Use when writing Blender Python code that must work across multiple versions (3.x/4.x/5.x), or when migrating scripts between Blender versions. Prevents breakage from renamed APIs, removed modules (bgl in 5.0), and changed function signatures. Provides the complete breaking changes matrix, deprecation timeline, and version-safe coding patterns. Keywords: Blender version, migration, deprecated, breaking change, bgl removal, bpy.app.version, version compatibility, 3.x, 4.x, 5.x, API changes, which Blender version, what changed in Blender 4.

GraphQL gives clients exactly the data they need - no more, no less. One endpoint, typed schema, introspection. But the flexibility that makes it powerful also makes it dangerous. Without proper controls, clients can craft queries that bring down your server.

Use when debugging Blender ReferenceError from removed objects, undo-invalidated bpy.data references, or stale ID pointers. Prevents the common crash of caching bpy.data references across undo/redo operations (references become invalid). Covers safe data access patterns, ID reference caching pitfalls, data lifecycle management, and bpy.data collections. Keywords: ReferenceError, undo, stale reference, bpy.data, ID pointer, data invalidation, removed object, data lifecycle, StructRNA, Blender crash, object disappeared after undo, StructRNA has been removed.

Use when debugging Blender AttributeError from removed APIs, ImportError from deprecated modules (bgl in 5.0), or breaking changes in operator signatures between versions. Prevents silent failures from using APIs that were renamed or removed in newer Blender versions. Provides migration patterns from 3.x to 4.x to 5.x with complete replacement mappings. Keywords: AttributeError, ImportError, deprecated, version error, bgl, migration, breaking change, removed API, 3.x to 4.x, 4.x to 5.x, Blender upgrade, module has no attribute, addon not compatible.

Use when automating Blender workflows -- batch rendering, headless processing via blender --background, file format conversion, or pipeline integration. Prevents the common mistake of using viewport operators in background mode (no UI context available). Covers command-line rendering, file I/O automation (OBJ/FBX/STL/USD/glTF), scene assembly, and CI/CD pipeline integration. Keywords: batch rendering, headless, blender --background, automation, file conversion, OBJ, FBX, STL, USD, glTF, pipeline, command-line, STEP import, CAD format, convert file format, export to web.

Use when generating mesh geometry for AEC applications -- buildings from vertices, IFC geometry visualization, parametric elements, or custom mesh tools. Prevents the performance mistake of creating vertices one-by-one instead of using from_pydata or BMesh batch operations. Covers mesh creation, BMesh algorithms, foreach_get/set optimization, and mesh analysis tools. Keywords: mesh generation, from_pydata, BMesh, parametric, building geometry, mesh analysis, foreach_get, foreach_set, vertices, AEC mesh, custom mesh tool, create mesh from code, generate 3D shape.

Use when building node trees via Python -- Geometry Nodes, Shader Nodes, or Compositor Nodes. Prevents the breaking change of using node.inputs/outputs by index instead of NodeTreeInterface (4.0+) for node group I/O. Covers node creation, linking, group management, and the NodeTreeInterface API migration. Keywords: node tree, Geometry Nodes, Shader Nodes, Compositor, NodeTreeInterface, node_group, links, nodes.new, node sockets, procedural geometry, create nodes from Python, connect nodes, add node.