jarmak-personal

PROACTIVELY USE THIS SKILL when the user says "commit", "land", "land this", "ship it", "done", "wrap up", "let's finish", or any intent to commit work. This is the ONLY user entrypoint for the commit workflow — do not invoke $pre-land-review directly for commits. Orchestrates the full landing flow: pre-land review, staging, review marker, and git commit.

PROACTIVELY USE THIS SKILL when wiring a new operation into the vibeSpatial Python dispatch stack — adding a public API method, connecting to GeometryArray, writing or updating the owned dispatch helper, handling CPU fallback observability, coercing GeometryArray or OwnedGeometryArray inputs, or extending DeviceGeometryArray surfaces. This is the Python-side complement to $new-kernel-checklist (which covers the kernel itself). Trigger on: "wire dispatch", "add API", "add method", "GeometryArray", "DGA method", "CPU fallback", "dispatch wrapper", "public API", "coerce", "OwnedGeometryArray".

The review gate that must pass before any commit lands. Called automatically by the $commit skill — do NOT invoke directly when the user says "commit", "land", "ship it", etc. (use $commit instead). Invoke directly only when you want to run the review without committing, or when another skill references it. This is a MANDATORY gate — do not create a git commit without completing this checklist.

PROACTIVELY USE THIS SKILL when wiring a new operation into the vibeSpatial Python dispatch stack — adding a public API method, connecting to GeometryArray, writing or updating the owned dispatch helper, handling CPU fallback observability, coercing GeometryArray or OwnedGeometryArray inputs, or extending DeviceGeometryArray surfaces. This is the Python-side complement to /new-kernel-checklist (which covers the kernel itself). Trigger on: "wire dispatch", "add API", "add method", "GeometryArray", "DGA method", "CPU fallback", "dispatch wrapper", "public API", "coerce", "OwnedGeometryArray".

PROACTIVELY USE THIS SKILL to find relevant files and documentation before starting any work. Use it instead of exploring the codebase manually. Trigger on: any new task, request, question, bug, feature, investigation, or when you need to find code, docs, tests, scripts, kernels, configs, ADRs, architecture docs, or understand how something works. Also trigger on: "where is", "find", "how does", "what file", "which module", "show me", "look up", "search for", "navigate to", "open", "locate", "explore", "investigate", "understand", "learn about", "read about", "check", "review", "audit", "debug", "fix", "modify", "update", "add", "implement", "create", "write", "build", "refactor", "test", "benchmark", "profile", "optimize". This is the fastest way to find anything in the repo — always prefer it over manual grep/glob exploration.

PROACTIVELY USE THIS SKILL when the user provides a PRD, spec, requirements doc, or tasklist and wants end-to-end execution with minimal check-ins. Trigger on: "autonomous mode", "own this", "own this end-to-end", "enact this PRD", "implement all of this", "here are the requirements", "no check-ins unless blocked", or similar. Treat the provided PRD or tasklist as the mandate. Execute through implementation, verification, profiling, docs, and landing when requested. Only interrupt for true external blockers such as missing secrets, required sandbox or network approval, destructive irreversible actions not already authorized, or contradictory requirements.

Use this skill to optimize existing CUDA/NVRTC kernel code, CuPy operations, CCCL primitive usage, or GPU dispatch logic in src/vibespatial/. Unlike gpu-code-review (which flags issues) and cuda-writing (which guides new code), this skill reads existing code and produces concrete rewrites with measured justification. Invoke on pre-existing kernel files to bring them up to NVIDIA best-practice performance standards.

PROACTIVELY USE THIS SKILL when writing, modifying, or reviewing GPU kernels, CUDA/NVRTC kernel source, CCCL primitive usage, device memory management, stream-based pipelining, or any GPU dispatch logic in src/vibespatial/. Covers kernel lifecycle, ADR-0033 tier system, stream overlap patterns, warp-level intrinsics, count-scatter patterns, precompilation (ADR-0034), precision dispatch (ADR-0002), and GPU saturation techniques.

PROACTIVELY USE THIS SKILL when writing, modifying, or reviewing GPU kernels, CUDA/NVRTC kernel source, CCCL primitive usage, device memory management, stream-based pipelining, or any GPU dispatch logic in src/vibespatial/raster/. Covers kernel lifecycle, CCCL primitives, shared-memory stencil tiling, CCL union-find patterns, cooperative primitives, GPU saturation techniques, and raster-specific dispatch patterns.

PROACTIVELY USE THIS SKILL when the user says "commit", "land", "land this", "ship it", "done", "wrap up", "let's finish", or any intent to commit work. This is the ONLY user entrypoint for the commit workflow — do not invoke /pre-land-review directly for commits. Orchestrates the full landing flow: pre-land review, staging, review marker, and git commit.

Use this skill to optimize existing CUDA/NVRTC kernel code, CuPy operations, CCCL primitive usage, or GPU dispatch logic in src/vibespatial/raster/. Unlike gpu-code-review (which flags issues) and gpu-kernel (which guides new code), this skill reads existing code and produces concrete rewrites with measured justification. Invoke on pre-existing kernel files to bring them up to NVIDIA best-practice performance standards.

PROACTIVELY USE THIS SKILL when reviewing GPU kernel code, CUDA/NVRTC source, CuPy operations, CCCL primitive usage, device memory management, stream-based pipelining, or any GPU dispatch logic. This skill contains quantitative thresholds, anti-pattern detection rules, and architecture-specific guidance for A100, H100, RTX 3090, and RTX 4090 GPUs. Use it to catch performance regressions, memory management issues, synchronization bugs, and access pattern problems before they land.

PROACTIVELY USE THIS SKILL when adding a new GPU kernel, NVRTC kernel, CCCL primitive wrapper, or any new GPU-dispatched operation to src/vibespatial/raster/. This checklist ensures every compilation, caching, dispatch, test, and documentation step is completed. Trigger on: "new kernel", "add kernel", "implement kernel", "write kernel", "create kernel", "new GPU operation", "add GPU", "scaffold kernel".

The review gate that must pass before any commit lands. Called automatically by the /commit skill — do NOT invoke directly when the user says "commit", "land", "ship it", etc. (use /commit instead). Invoke directly only when you want to run the review without committing, or when another skill references it. This is a MANDATORY gate — do not create a git commit without completing this checklist.

Use this skill to verify raster processing correctness, check nodata semantics, validate affine transforms, understand neighborhood operations, or answer domain questions about raster algorithms. This is the domain knowledge oracle for autonomous agents working on vibespatial-raster. Trigger on: "is this correct", "should this return", "nodata", "affine", "CRS", "neighborhood", "stencil", "focal", "morphology", "CCL", "connected component", "zonal", "rasterize", "polygonize", "marching squares".

Translates task requirements into GitHub Copilot CLI commands. Used by copilot-driver agent to execute coding tasks via Copilot.

Translates task requirements into Cursor CLI commands. Used by cursor-driver agent to execute coding tasks via Cursor.

Intelligently routes and delegates coding tasks to the best AI agent (Codex, Cursor, Gemini, Copilot). Use /delegate <task> to automatically pick the optimal agent based on task type and complexity.

Translates task requirements into Gemini CLI commands. Used by gemini-driver agent to execute coding tasks via Google Gemini.

"Structured brainstorming with research-driven exclusion zones and shape uniqueness enforcement. Forces divergent thinking through volume and a hard constraint: no idea can take the same shape as any prior art or any previous idea. Use when exploring solutions to a problem, designing features, or when the obvious approach might not be the best one."

Translates task requirements into Codex CLI commands. Used by codex-driver agent to execute coding tasks via OpenAI Codex.

PROACTIVELY USE THIS SKILL when the user says "commit", "land", "land this", "ship it", "done", "wrap up", "let's finish", or any intent to commit work. This is the ONLY user entrypoint for the commit workflow — do not invoke /pre-land-review directly for commits. Orchestrates the full landing flow: pre-land review, staging, review marker, and git commit.

Use this skill to optimize existing fused NVRTC kernel code, GPU dispatch logic, or the Helmert datum shift kernel in src/vibeproj/. Reads existing code and produces concrete rewrites with measured justification. Invoke on pre-existing kernel code to bring it up to NVIDIA best-practice performance standards.

PROACTIVELY USE THIS SKILL when writing, modifying, or reviewing GPU kernels, fused NVRTC kernel source, or any GPU code in src/vibeproj/. Covers fused kernel architecture, preamble/postamble system, precision parameterization ({real_t}), device helper functions, kernel compilation via CuPy RawKernel, the Helmert datum shift kernel, and double-single arithmetic.

PROACTIVELY USE THIS SKILL when reviewing GPU code changes in vibeProj. Provides the 6-pass review procedure with thresholds, anti-patterns, and hardware specs specific to fused NVRTC projection kernels.

The review gate that must pass before any commit lands. Called automatically by the /commit skill — do NOT invoke directly when the user says "commit", "land", "ship it", etc. (use /commit instead). Invoke directly only when you want to run the review without committing, or when another skill references it. This is a MANDATORY gate — do not create a git commit without completing this checklist.

Run a repeatable black-box Skillager setup and working-readiness simulation in a fresh temp directory and report discovery, scoped exposure decisions, and UX findings.

Run Dungeoneer system simulation scenarios with fresh low-context agents, clean workspace snapshots, structured YAML artifacts, runtime traces, and optional persistence to ../dungeoneer-sim-history. Use when asked to run, rerun, compare, audit, or save Dungeoneer simulation runs, especially fresh simulator runs for scenario YAML files under docs/simulations/scenarios.

User-facing control skill for vibe-while-u-vibe. Use when the user says things like "/vibe stop", "/vibe louder", "/vibe skip", "play focus music", "what's playing", "where are my tracks", "/vibe disable", or "/vibe uninstall".