AI & ML

loulanyue Skill

Decision framework for choosing between regex and LLM when parsing structured text — start with regex, add LLM only for low-confidence edge cases.

loulanyue Skill

选择在解析结构化文本时使用正则表达式还是大型语言模型的决策框架——从正则表达式开始，仅在低置信度的边缘情况下添加大型语言模型。

loulanyue Skill

Decision framework for choosing between regex and LLM when parsing structured text — start with regex, add LLM only for low-confidence edge cases.

loulanyue Skill

Apple FoundationModels framework for on-device LLM — text generation, guided generation with @Generable, tool calling, and snapshot streaming in iOS 26+.

hojinjava Bundle

Use when starting RAG on a new corpus and the chunking/loading design is not yet decided - runs as a team (one orchestrator dispatching measurer, reader, refuter, checker agents in parallel rounds), loads the corpus into a local DuckDB and measures the raw documents first, then picks chunking, enrichment, loading scope, metadata filters and cross-corpus links from evidence instead of defaults. The deliverable is a CONSULTING presentation, not a measurement log - its table of contents is the argument itself and every number traces back to a query. Scope is the corpus axis only: retrieval method, query transformation and post-processing need an evaluation set and stay unresolved here. Triggers on "RAG 프로파일링", "코퍼스 프로파일링", "이 데이터로 청킹 어떻게", "임베딩 대상 정해야 함", "코퍼스 분석부터", "어떤 RAG 기법 써야 하나", profiling a document folder before indexing.

hojinjava Skill

Use when scoring a coding model or coding agent against a rubric from its real output - PRs, branches, diffs, generated docs - and the score has to survive being challenged. Runs objective gates before any human judgment, requires the evaluator to build and run the code with the exact commands the task instructions specified, and runs the same prompts blind through a control model so a deduction can be attributed to the model rather than to a broken instruction. Triggers on "모델 평가", "성적서", "채점 기준", "vibe eval", "코딩 모델 비교", "에이전트 벤치마크", "PR 기준으로 평가", "대조군", scoring an AI coding tool trial or vendor bake-off.

l0kyurue1 Bundle

Use when the desired outcome is a user-facing Markdown preview or editor: the user wants to inspect rendered output, edit the document themselves, or receive it in an artifact-like interface. Also use for a user-facing interface for installed agent skills, agent configuration, or memory. Stay quiet when Markdown only needs internal agent work such as reading, checking, reviewing, or editing.

swaylq Bundle

说人话.skill — 把难懂、别扭、不通顺的中文改成一读就懂的人话，意思一个字不变。双外脑改写：GPT-4o 和 Gemini（走 OpenRouter）各改一版，本地 agent 对照原文逐句查「意思有没有跑」，合成终稿；原文永远保留。两个固定时机：备料（写作素材先洗一遍）、出稿（交付前最后捋一遍）；也可以单独对任何一段文字用。首次使用需要用户配一个 OpenRouter key。 触发词：「说人话」「这段话看不懂」「帮我捋顺这段」「太绕了」「翻译成人话」「大白话讲一下」「备料清洗」「出稿前过一遍」「读不通」「翻译腔太重」。

tomsej Skill

Review a change with an optional workflow contract using deterministic scope, adaptive cross-model reviewers, verified findings, fixes, and a gate. Use standalone or from wf-impl.

barastrong Bundle

Use when working on server-side code, REST/GraphQL APIs, database queries, authentication/authorization, background jobs, microservices, or any task involving Node.js, Python (Django/Flask/FastAPI), Go, Java (Spring), Rust, or .NET backend. Trigger phrases: "build API", "server endpoint", "database migration", "backend logic", "service layer", "middleware", "ORM model", "queue worker", "caching strategy".

awesome-ai-garage Skill

Conducts multi-axis code review. Use before merging any change. Use when reviewing code written by yourself, another agent, or a human. Use when you need to assess code quality across multiple dimensions before it enters the main branch.

gethlx Bundle

太乙枢机个人 Agent Skill。用于中医经典原文、概念与跨书关系解释，一味或多味组成、无方名方剂和方剂比较的条件性分析，个人病例事实记录与中医辨证，明确激活病例的同案复察，以及用户明确要求后的同源患者向移动端 H5 或纯图片可视化报告；当用户要求经典研析、方义方证、病例记录或分析、事实纠正、治疗反应比较、Day 1/2/3 观察或只读可视化报告时使用。

ahmadarifff Bundle

UI/UX design-system intelligence - 84 styles, 192 palettes, 74 font pairings, 25 charts, 99 UX guidelines across 22 stacks (searchable dataset + CLI). Internal genjutsu module: loaded by /genjutsu:cast and /genjutsu:paint, not invoked directly.

s0ld13rr Skill

SSH version CVEs and post-auth tricks — regreSSHion, user enum, agent hijack. Use when SSH is open. Triggers - port 22, OpenSSH banner, regreSSHion CVE-2024-6387, CVE-2018-15473.

aoyunyang Bundle

Pure-web protocol reverse skill: turn hostile browser clients into browser-free Python collectors. Auto-judge intake and tool path from signals: artifact-only first when samples suffice; for a fresh web live-target, collect sequential fingerprint-baseline then debugger-trace evidence (default chrome-devtools then js-reverse; upgrade baseline host to Camoufox/managed profile only on fingerprint pressure); for continuation, reuse the current gate when target and environment are unchanged. Route only mounted web-relevant MCP families. Out of primary scope: APK, native app, and mini-program reverse as the main path. Use for hostile web sign, token, cookie, WebSocket, GraphQL, protobuf, response-decode, browser-fingerprint, WebAssembly, challenge-bootstrap, dynamic-font, or protocol-collector flows.

samin12 Bundle

Use, configure, theme, extend, and orchestrate Hermes Agent.

samin12 Bundle

Use when the user wants to connect to Alpaca, inspect Alpaca account or market data, or place trades through the Alpaca MCP server.

samin12 Bundle

Parallel 4-agent cleanup of recent code changes.

samin12 Bundle

Create and manage generative game zones for Claude Command Center's Pokemon-style world. Use this skill when creating, updating, or designing game worlds via MCP tools.

samin12 Bundle

This skill should be used when the user asks to "create a persona", "add a new persona", "set up a new agent role", "add a woterclip persona", or runs the /persona-create command. Interactively creates a new WoterClip persona with SOUL.md, TOOLS.md, and config.yaml.

samin12 Bundle

This skill should be used when the user asks to "import a paperclip agent", "convert paperclip to woterclip", "migrate from paperclip", "import persona from paperclip", or wants to convert existing Paperclip agent directories into WoterClip persona format.

samin12 Bundle

Generate images, video, and audio via diffusion workflows.

samin12 Bundle

HuggingFace hf CLI: search/download/upload models, datasets.

samin12 Bundle

Inspect a customer agent codebase for its tool-management approach and framework, fetch Ratel docs, and write a markdown plan to integrate Ratel — integration mode (direct SDK / Ratel Local / hybrid), pilot scope, A/B test design, and rollout metrics keyed off Ratel's native OTLP telemetry. Use to add Ratel, set up the SDK, add tool search / hybrid retrieval, plan a Ratel pilot, rollout, or A/B test, or `/ratel-integrate`. Writes to repo/.ratel/, asks when ambiguous, never edits code; runs after /ratel-observability-assessment.

samin12 Bundle

The Eden MCP exposes one user's Eden Canvas account: their **workspace** (boards, notes, saved cards), the **social-intelligence** layer behind Eden Discover (cross-platform creator analysis + indexed post search), and the **post scheduler** (draft → schedule → publish across X, Threads, LinkedIn, Substack, Instagram, TikTok, Facebook, YouTube).

samin12 Bundle

Karpathy's LLM Wiki: build/query interlinked markdown KB.

samin12 Bundle

This skill should be used when the user asks to "check woterclip status", "show agent status", "what is woterclip doing", "show heartbeat status", "what's in the queue", or runs the /woterclip-status command. Shows current WoterClip state, issue queue, and blocked items.

samin12 Bundle

This skill should be used when the user asks to "run a heartbeat", "run the agent loop", "process Linear issues", "check for work", or runs the /heartbeat command. Executes the WoterClip heartbeat — picks up Linear issues, resolves personas, does work, and reports back.

samin12 Bundle

vLLM: high-throughput LLM serving, OpenAI API, quantization.

samin12 Bundle

Control TouchDesigner via twozero MCP.

samin12 Bundle

W&B: log ML experiments, sweeps, model registry, dashboards.

dimon-ton Bundle

Create, revise, audit, or prompt character sheets that preserve a character's identity across many outfits, poses, scenes, and image-generation models. Use when the user asks for a reusable character sheet, identity bible, wardrobe or outfit system, clothing item library, prompt adapter, expression/pose sheet, consistent character prompts, or help separating character identity from clothing and styling.

nurfarazi Skill

Planning-only orchestrator for bugs and features. Grills the user with clarifying questions, fans out parallel explorer agents to map the codebase, checks available skills, and produces a phased implementation plan with acceptance criteria saved as a .md file. Use when the user asks to plan a feature, plan a bug fix, create an implementation plan, or says "plan this" / "make a plan". This skill NEVER writes or edits code.

nurfarazi Skill

Execution orchestrator that implements a phased plan (typically from docs/plans/, produced by agent-orchestrator-planner) using dedicated subagents per phase. The orchestrator never implements code itself — it delegates implementation and progress tracking to specialized agents, and runs a single code review at the end (Cross-Validation Phase) against the full branch diff, and keeps the plan document updated with live progress. Use when the user asks to execute a plan, implement a plan, run the plan, or continue executing a phased plan.

zenx0x Bundle

Execute an approved Research Routes context across software, research-exploration, or hybrid journeys with concurrent route owners, shared control-plane resources, CounterPilot read-only boundaries, explicit evidence and provenance, lifecycle recovery, and HumanDecision seams. Use when implementation or experiments must proceed without treating AI inference as fact or silently changing canonical state.

zenx0x Bundle

Plan software, research-exploration, or hybrid work from a repository, question, problem, frame, or arbitrary starting point. Use when Codex needs a route-neutral map, shared backbone, concurrent routes and branches, explicit claims and hypotheses, provenance, evidence polarity, lifecycle recovery, or a plan that keeps candidate inference separate from HumanDecision and canonical state.