AI & ML

agentic-in Bundle

Serves LLMs with high throughput using vLLM's PagedAttention and continuous batching. Use when deploying production LLM APIs, optimizing inference latency/throughput, or serving models with limited GPU memory. Supports OpenAI-compatible endpoints, quantization (GPTQ/AWQ/FP8), and tensor parallelism.

agentic-in Bundle

OpenAI's general-purpose speech recognition model. Supports 99 languages, transcription, translation to English, and language identification. Six model sizes from tiny (39M params) to large (1550M params). Use for speech-to-text, podcast transcription, or multilingual audio processing. Best for robust, multilingual ASR.

saml Skill

Selects what models to use for various types of work.

saml Skill

Updates agent memory files. Trigger this automatically when agent memory needs to be updated.

saml Skill

Implements from the latest plan file using a cheap sub-agent, then review-fix loops until approved.

aqwsde321 Bundle

Create a redacted, portable handoff document that lets a fresh agent or session resume the current conversation without replaying the transcript. Use only when explicitly invoked as $handoff or when an active workflow explicitly requires a cross-session or cross-agent transfer; do not use for an ordinary summary or native Codex task transfer.

aqwsde321 Bundle

Build and sharpen a project's domain model. Invoke through $domain-modeling when the user wants to pin down domain terminology or a ubiquitous language, record an architectural decision, or another skill explicitly delegates glossary or ADR maintenance. Do not use for source-grounded project-context documentation.

jwiegley Skill

Multi-phase, multi-model deep analysis workflow for complex problems. This skill should be used when the user wants rigorous, multi-model collaborative analysis: deep research with Fable/Opus and PAL MCP consensus (GPT-5.5-Pro + Gemini 3 Pro), strategic planning, Fable/Opus execution with tests, comprehensive review, and adversarial devil's advocate critique. Invoke explicitly with /forge.

jwiegley Skill

Compress and simplify prompts to preserve meaning while reducing use of context. Use when asked to compress, shorten, or "caveman" a prompt or other text, or when text must fit a smaller context budget without losing meaning.

lingwei-zheng Bundle

Create or edit native Draw.io diagrams and optionally export editable PNG/SVG/PDF files. Use when the user explicitly requests Draw.io or an editable diagram, or when engineering-figure-agent selects Draw.io for a reviewed, label-heavy conceptual framework, mechanism diagram, architecture, workflow, ER diagram, sequence diagram, network diagram, or other topology-sensitive figure. For unreviewed academic conceptual figures, use conceptual-figure-workflow first; use nature-figure for statistical plots and maps.

lingwei-zheng Bundle

Use only when the user explicitly asks to hand off, compact, transfer, or summarize the current conversation for a fresh agent or later session. Create a concise temporary handoff document that points to existing artifacts, records unresolved work, and recommends relevant skills. Use sync instead for durable cross-device research-project state, directory initialization, or ongoing analysis and manuscript logs.

lingwei-zheng Bundle

Use when the user needs to execute an already scoped engineering or research-paper figure brief, including system architecture diagrams, algorithm workflows, hardware schematics, mechanism diagrams, graphical abstracts, editable Draw.io diagrams, redraws, or image edits. Uses the official OpenAI image path by default, with Gemini, Nano Banana, and other providers retained as explicit opt-in compatibility paths. For conceptual framework or experiment-design figures whose theory, variables, mechanisms, or claim-evidence boundaries are not yet reviewed, use conceptual-figure-workflow first.

lingwei-zheng Bundle

Use when the user wants a conceptual framework figure, experiment design diagram, thesis framework, research mechanism diagram, theoretical model figure, variable relationship diagram, or paper overview schematic. This skill first uses academic-research to review the theory structure, research question, variables, mechanisms, hypotheses, and claim-evidence boundaries, then produces a figure brief for engineering-figure-agent to execute.

gurudevvv Bundle

Structured deep research on any topic. Asks clarifying questions before starting, sets success criteria, runs multi-agent research with web search, rates confidence, delivers a concrete decision, and saves the artifact. Use when: user needs thorough research ("исследуй", "разберись в", "deep research", "найди всё про", "помоги разобраться", "проанализируй тему", "research", "что лучше выбрать", "сравни варианты", "нужен анализ"). Do NOT use for: quick factual questions, code debugging, project setup (use spec-first), design tasks (use design-first), code review, or tasks where the answer is obvious.

gavishap Skill

The relay's synchronous mode — make Claude and Codex argue BEFORE you build, inside one session, with zero copy-paste. Four phases — SCOUT (recon from the vault's layered memory), LOCK (interrogate the user until intent is locked), SPAR (the rival model attacks the plan in bounded rounds, read-only), SHIP (one model builds, the other grades the diff, then the argument is compiled into the Wiki). Use when the user says "/spar", "spar this", "have codex review this plan", "argue this with codex", "cross-model review", "stress-test this plan", "codex build this", "have the models fight", or is about to build something high-stakes (auth, schema, migrations, concurrency, payments, greenfield architecture). NOT for trivial changes, and not a substitute for the async relay baton (that's the `relay` skill).

gavishap Skill

Validate Wiki and source note health. Use before commits, after ingests, or when notes were edited by hand.

totvs Skill

Gera o esqueleto de um Form (formulário eletrônico) do Fluig — definição de campos, view/markup, eventos e (quando solicitado) tabela Pai x Filho com tablename, template de linha e orientações de consumo via dataset, aplicando as convenções oficiais. Use quando o desenvolvedor pedir para criar/iniciar um novo formulário do Fluig a partir de um nome ou propósito e seus campos.

totvs Skill

Gera o esqueleto de um Dataset customizado do Fluig — o arquivo JavaScript server-side com a(s) função(ões) nomeada(s) que a plataforma invoca para construir e retornar o conjunto de dados via API pública de Dataset, aplicando filtros/constraints e ordenação. Use quando o desenvolvedor pedir para criar/iniciar um novo dataset customizado a partir de um nome/propósito, campos e fontes de dados.

vikast908 Bundle

Help the user critique and improve interfaces using Norman's interaction principles. Ground every recommendation in named principles from The Design of Everyday Things by Don Norman (cite concept + section only if present in references/book-passages.md). Use when the user runs /doet or /design-everyday-things, or discusses: UX critique, affordance, Norman, usability, mental model, error design.

xiwan Bundle

AI Development Lifecycle — standardized, version-driven workflow for AI agents developing on codebases. Two modes: (1) Assisted: human drives each phase decision. (2) Autonomous: agent self-directs (declare in task context). On first use, agent analyzes project and generates OPERATIONS.md at root. Each task follows 7 phases: orientation (mandatory summary), design, scratch validation, implementation, testing, documentation, commit. Progress tracked in version log. Use when starting any coding task, refactoring, or bugfix on a codebase.

xiwan Bundle

Translate PowerPoint (.pptx) files between languages using Amazon Translate or Bedrock LLM. Use when: (1) translating presentations to other languages, (2) localizing slide decks for international audiences, (3) converting Chinese/Japanese/Korean presentations to English or vice versa. Supports two engines: Amazon Translate (fast, cost-effective) and Bedrock LLM (context-aware, better quality for nuanced content). Preserves formatting, styles, and layout.

xiwan Skill

Game design coach — guide users from a one-line idea to a high-quality OpenGame prompt step by step. Trigger when user says "design a game", "game idea", "help me think of a game", "game design", "refine prompt", etc.

xiwan Bundle

像素游戏美术风格方法论与质量控制框架。用于：(1) 定义游戏美术风格规范（分辨率、调色板、光照方向等）， (2) 指导 AI 生成像素美术素材时的 prompt 约束和后处理流程， (3) 角色/场景/UI/特效的设计规范和验收标准， (4) 外包美术素材的风格统一验收。 触发词：像素美术、pixel art、游戏美术风格、美术规范、style guide、调色板、tilemap、sprite、角色动画、 The Last Night、2.5D 电影感像素、filmic pixel art。

mxrsv Skill

Compact the current conversation into a handoff document for another agent to pick up.

mxrsv Skill

Use to audit token consumption across always-loaded guidance — agents, skills, MCP servers, CLAUDE.md / AGENTS.md, rules, hooks and config — in Claude Code or Codex. Identifies bloat, redundancy and duplicate skill roots, and produces optimization recommendations. Meta-maintenance skill.

sarveshtalele Bundle

One or two sentences describing what the skill does, followed by trigger phrases the agent should recognize. Trigger when the user says: ...

sarveshtalele Bundle

Turn an approved specification into a technical plan (plan.md): architecture, components, data model, key decisions, and milestones. Trigger on: create a plan, technical plan, implementation plan, SDD plan, design from spec.

agentic-in Bundle

Guarantee valid JSON/XML/code structure during generation, use Pydantic models for type-safe outputs, support local models (Transformers, vLLM), and maximize inference speed with Outlines - dottxt.ai's structured generation library

agentic-in Bundle

Foundation model for image segmentation with zero-shot transfer. Use when you need to segment any object in images using points, boxes, or masks as prompts, or automatically generate all object masks in an image.

agentic-in Bundle

State-of-the-art text-to-image generation with Stable Diffusion models via HuggingFace Diffusers. Use when generating images from text prompts, performing image-to-image translation, inpainting, or building custom diffusion pipelines.

agentic-in Bundle

Track ML experiments with automatic logging, visualize training in real-time, optimize hyperparameters with sweeps, and manage model registry with W&B - collaborative MLOps platform

agentic-in Bundle

Evaluates LLMs across 60+ academic benchmarks (MMLU, HumanEval, GSM8K, TruthfulQA, HellaSwag). Use when benchmarking model quality, comparing models, reporting academic results, or tracking training progress. Industry standard used by EleutherAI, HuggingFace, and major labs. Supports HuggingFace, vLLM, APIs.

awaik Bundle

Работать с Авито Рекламой через LidFly MCP v3: кабинеты, кампании, группы, бюджеты, статистика, доступы и юридические операции. Использовать для аудита или управления Avito Ads с проверкой 9-значного account_id и безопасным write workflow.

awaik Bundle

Работать с обычным профилем Авито через LidFly MCP v3: объявления, Autoload/автозагрузка, сообщения, отзывы, заказы, остатки, продвижение и доступные партнёрские API. Использовать для задач Авито вне отдельного рекламного кабинета «Авито Реклама».

awaik Bundle

Анализировать Яндекс Метрику через LidFly MCP v3: счётчики, цели, UTM, Директ, CPA, конверсии, страницы и сравнение периодов. Использовать для отчётов и диагностики с точным counter_id и без client_login.

awaik Bundle

Разрешать provider scope в LidFly MCP v3 через get_provider_context и resolve_campaign_scope. Использовать вместе с provider-скиллом, когда кабинет, подключение, кампания или workspace_project_id не заданы точно либо пользователь называет объект по имени.