Provides guidance for writing and benchmarking optimized CUDA kernels for NVIDIA GPUs (H100, A100, T4) targeting HuggingFace diffusers and transformers libraries. Supports models like LTX-Video, Stable Diffusion, LLaMA, Mistral, Qwen, and Qwen3. Includes integration with HuggingFace Kernels Hub (get_kernel) for loading pre-compiled kernels. Includes benchmarking scripts to compare kernel performance against baseline implementations.

Run an external LLM code review with Codex CLI, Gemini CLI, or both. Use when the user asks for a second opinion, external review, Codex review, Gemini review, or wants a model-vs-model review of current changes, a branch diff, a specific commit, or a GitHub pull request.

Generates minimal macOS Seatbelt sandbox configurations. Use when sandboxing, isolating, or restricting macOS applications with allowlist-based profiles.

Find and analyze screenshots from ~/Downloads/Screenshots/ for UI/UX review. Use when asked to look at screenshots, review UI designs, inspect interface errors, analyze layout issues, or examine visual problems. Triggers on "check my latest screenshot", "review my UI", "analyze this screen capture".

Save AI agent session information for later resumption. Use when: (1) Finishing a work session and want to save progress (2) User says "save session", "remember this session", or "save your session ID" (3) Before ending a long coding/debugging session (4) When user wants to be able to resume this conversation later Supports Claude, Codex, Gemini, Copilot, Cursor, and other AI agents.

Profile Rust code using samply to identify CPU bottlenecks. Use when performance is slow, before optimizing, or when the user asks to profile.

Build terminal UIs with ratatui following 2026 Rust best practices. Use when: (1) Creating new TUI apps, (2) Adding widgets/layouts, (3) Keyboard navigation/state management, (4) Image integration via ratatui-image, (5) Async event handling, (6) Release optimization. Covers v0.30.0+ API, Elm Architecture, StatefulWidget, color-eyre.

Orchestrate multi-step work with Claude Code CLI as a bounded worker while Codex stays lead developer. Use when Codex should split a task into scoped sub-tasks, delegate planning or implementation to Claude with `claude -p`, run parallel Claude sessions, isolate risky edits in worktrees, define custom Claude subagents, or move repeated enforcement into Claude hooks/settings.

Behavioral checklist for disciplined LLM coding. Use when starting non-trivial tasks to enforce assumption surfacing, minimal changes, and verifiable goals. Derived from Andrej Karpathy's observations on LLM coding pitfalls.

Use when profiling native macOS or iOS apps with Instruments/xctrace. Covers correct binary selection, CLI arguments, exports, and common gotchas.

Remove signs of AI-generated writing from text. Use when editing or reviewing text to make it sound more natural and human-written. Based on Wikipedia's comprehensive "Signs of AI writing" guide. Detects and fixes patterns including: inflated symbolism, promotional language, superficial -ing analyses, vague attributions, em dash overuse, rule of three, AI vocabulary words, negative parallelisms, and excessive conjunctive phrases. 30c5c8d (Update humanizer plugin to upstream v2.2.0)

Produce a single self-contained .html artifact (no build, no external assets) when the task is spatial, interactive, visual, multi-axis, or long-running. Triggers on "build me an HTML for X", "make a dashboard / status snapshot / comparison / decision tree / timeline / architecture map", "render this plan as a clickable site", or when content can't fit markdown structurally. Skips short replies, code-only outputs, simple plans.

Generate optimized handoff prompts for delegating work to another LLM agent. Use when handing work to GPT-5.x/Codex, Claude 4.x, Gemini 3.x, or Grok 4.x, either as a shared-workspace sub-task handoff or a fresh-context handoff for a new session or model. Triggers on requests like "create a handoff prompt", "delegate this task to another agent", "hand this off", or "prepare context for another agent".

Measure and improve Go program performance on modern Go (1.24+). Use when profiling Go code, diagnosing CPU or memory bottlenecks, investigating latency or contention, writing or fixing benchmarks, comparing benchmark results, using pprof or trace data, applying PGO, or tuning hot-path Go code.

Simplifies and refines code for clarity, consistency, and maintainability while preserving all functionality. Focuses on recently modified code unless instructed otherwise.

Deliberate fresh-eyes self-review and repair after making changes. Use when an agent has just written or modified code, config, tests, or docs and should pause to look for obvious bugs, regressions, missing tests, confusing behavior, or risky assumptions, fix the clear local issues it finds, and only then finalize, hand off, or commit. Supports Claude Code, Codex, and Gemini with provider notes in references/.

Build, sign, and deploy iOS/macOS apps using Zed Editor and CLI-only workflows without Xcode UI. Use when: (1) Setting up a new iOS/macOS project with XcodeGen, (2) Configuring secure credential management via xcconfig files, (3) Setting up Zed Editor with Swift LSP, (4) Building/deploying to physical devices via CLI, (5) Creating CI/CD pipelines for Apple apps, (6) Troubleshooting CLI build issues. Covers XcodeGen, xcode-build-server, justfile commands, devicectl, and GitHub Actions scaffolding following 2026 best practices.

Search and retrieve research paper metadata using the Semantic Scholar Academic Graph API via curl. Use this skill whenever the user wants to find academic papers, look up citations, get paper details by DOI/arXiv ID/title, explore an author's publications, or fetch reference/citation lists. Trigger on phrases like "find papers on X", "look up this paper", "how many citations does X have", "papers citing X", "search for research about X", "get metadata for arxiv:...", or any request to explore academic literature. Always use this skill when the task involves academic paper search or metadata retrieval — even if the user just pastes a DOI or arXiv link and wants info about it.

Search and retrieve scholarly metadata from the OpenAlex API — a free, open catalog of 270M+ works, 90M+ authors, and 100K+ sources. Use this skill whenever the user wants to query OpenAlex for works, authors, institutions, sources, topics, publishers, or funders. Trigger on phrases like "search OpenAlex for X", "find papers in OpenAlex", "OpenAlex works by author Y", "get institution metadata from OpenAlex", "look up this DOI in OpenAlex", "how many works does institution X have in OpenAlex", or any request that specifically involves the OpenAlex API or database. Also trigger when the user pastes an OpenAlex ID (like W1234567890 or A5023888391) or mentions OpenAlex by name in any research context. This skill complements the semantic-scholar skill — use OpenAlex when the user asks for it specifically, when they need institution/funder/topic data that Semantic Scholar doesn't cover, or when they want open-access filtering and aggregation features unique to OpenAlex.

Use this skill when creating or editing a topic map in the Keine knowledge base. Trigger after ingesting a set of related entries on a topic, or when asked to summarize, compare, survey, or give an overview of a subject area. Also use when a user asks a broad question and you want to leave a synthesized overview for future agents or readers. Maps are different from entries — they synthesize rather than document.

Use this skill when creating a new knowledge entry or editing an existing one in the Keine knowledge base. Trigger when asked to add, ingest, document, or update any piece of knowledge — a URL, PDF, book, concept, note, or research finding. Use this skill even when the user doesn't say "entry" — if they want to capture or update knowledge, this is the skill.

Use this skill only when the user explicitly requests deep research or a detailed report on a topic — phrases like "research X", "give me a detailed report on X", "deep dive into X", "I want a thorough analysis of X", or "write me a research report on X". Do NOT trigger for casual questions, quick lookups, or requests to add/edit entries. This is a heavyweight, multi-step workflow that mines the knowledge base, fills gaps with web research, creates new KB entries along the way, and produces a long-form, citation-rich research report saved to reports/.

Always use this skill before creating, editing, or tagging any document in the knowledge base. Use it when asked to add, ingest, find, link, or manage any entry.

Use this skill whenever the user wants to discuss, explore, or ask questions about a topic using the local knowledge base as grounding. Trigger on phrases like "what do you know about X", "let's talk about X", "explain X", "I'm trying to understand X", "discuss X with me", "what does the KB have on X", "help me think through X", "tell me about X", or any conversational question about a concept or idea. Also trigger when the user asks a question that could be answered from stored knowledge, even if they don't explicitly say "knowledge base". This is the default skill for knowledge-grounded conversation — prefer it over answering from memory alone whenever the KB might have relevant content. Do NOT trigger for: deep research reports (use keine-research), adding new entries (use keine-update-entries), or creating maps (use keine-update-maps).