AI & ML

dprakash2101 Skill

Threat-model, implement, or review security-sensitive .NET and ASP.NET Core behavior. Use for authentication, authorization, secrets, untrusted input, uploads, outbound URLs, cryptography, or an explicit security review; not for routine style review.

lingyu9495-source Bundle

平台现在不是"猜"你是不是 AI 写的，是**拿语义指纹比对**——句子节奏、关联词习惯、段落结构、标点规范度，都对得上就降权。 问题是：**你根本不知道自己的稿子哪里像 AI**，只感觉"读起来有点平"。 这套做的是**检测 → 改写 → 复检**的闭环，不是丢你一段 prompt： 1. **五维指纹检测**（可定位到句）：句式指纹 / 词汇指纹 / 结构指纹 / 节奏指纹 / 标点指纹，逐维打分并**指出最像 AI 的那几句**——先知道病在哪，才谈得上改 2. **分层改写手法**（不是换词）：具体化替换（抽象论断 → 具体事实/数字/场景）、节奏打散（破坏均匀句长）、视角注入（第一人称经验与时间地点）、信息增量（补真实案例）——**平台真正奖励的是"人工介入痕迹"，不是同义词替换** 3. **闭环复检**：改完自动复检，未达标按最弱维度再改一轮（可设最多轮次），并输出前后对比报告 4. **平台差异**：头条（语义指纹比对 + 原创度权重）、公众号（完读率第一）、小红书（AI 辅助需标注）、网文（平台审核红线）各有侧重 ⚠️ **合规边界写死在包里**：这不是"洗稿工具"，也不是"绕过审核"——是把你自己的内容补足人工痕迹（真实案例、实地数据、个人观点），这恰好是平台规则里明确要求的方向。用在别人的稿子上＝洗稿，包内直接标红。 输入：你的原稿（.md / .txt / 直接粘贴） 输出：五维指纹报告（含问题句定位）+ 改写建议 + 改写后复检对比 触发词：AI味、去AI味、降AI率、AI指纹、AI检测、内容原创度、平台降权、AI生成内容、人工痕迹、改写、去AI化、过检测。

lingyu9495-source Bundle

你给 AI 助手派了个重活，然后发现：发消息它不回、想插话只能排队、等十几分钟才回一句——等它抬头，需求早变了。根因不是模型慢，是主脑把重活放在前台跑，整个回合被占死。这个 Skill 把这套「重活外派 + 随时应答」机制做成可落地的东西：① 规则块（一页写清判级、派完立刻让出对话、工单四要素、只回收结论、自报≠事实、拍板留主脑、跨会话任务换机制）② 跨平台落地脚本 provision_agent_responsiveness.py：探测本机已有的 Agent 环境（Hermes 的 SOUL 置顶铁律块、Claude Code 的全局指令文件，这两条已实测），把规则块幂等写入（自动备份原文件、重复执行不重复写、绝不覆盖你已有内容）；其余平台（Codex / Cursor / WorkBuddy / 任意 Agent）则导出一份可直接粘贴的规则块 markdown，你粘到系统提示或自定义指令里就生效 ③ 体检脚本 audit_agent_responsiveness.py：逐项核对规则块是否存在、是否被后续改动覆盖、探测到哪些平台——没检测到就如实报「未检测到支持的平台」，不假报通过。纯标准库、零依赖、Windows/macOS/Linux 通用。触发词：AI不回话、等太久、插话卡住、助手响应慢、主脑被占住、重活外派、子代理、多智能体、响应速度、落地配置、系统提示。

thinslatelabs Skill

Use when a user wants to build or manage their Keepp page — adding, arranging, or removing blocks (links, cards, products, bookings, forms, maps, YouTube videos, menus, tickers, profile, socials, headers, text), scheduling a block to appear later, changing the layout, or restyling the theme — through the Keepp Agent API. Requires a Keepp Pro API key (keepp_live_…).

raunakkathuria Bundle

Audit existing software for closed-set semantic decisions that TypeSafe's Jev model could replace or harden. Use when the user asks where Jev, TypeSafe, or System One fits, or asks to inspect prompt-and-parse flows, LLM judges or classifiers, semantic routing, manual triage, or unmeasured heuristic thresholds. Do not use for general AI ideation unless the user is asking about decision automation in an existing system. Produce a plain-language report with evidence, actionable steps, explicit non-fits, and checked request bodies.

zhangweildlh Bundle

通过 Playwright（含 MCP 浏览器工具）实现浏览器自动化：导航网页、点击元素、填写表单、截图、提取数据、调试真实浏览器工作流。适用场景：(1) 需要真实浏览器而非静态抓取时；(2) 任务涉及 Playwright MCP、浏览器工具、Playwright 测试/脚本或 JS 渲染页面时；(3) 用户希望将导航、表单、截图、PDF、下载或浏览器驱动的提取转化为可靠结果时。

khe618 Skill

Append durable findings — non-obvious bug fixes, tooling quirks, counterintuitive library/API behavior, anything that took real time to figure out and will plausibly bite again — to a two-tier log. Tier 1 (project-specific findings) goes to a per-subproject `LEARNINGS.md`; tier 2 (cross-project findings about Claude Code itself, OS/shell tooling, or third-party libraries) goes to a topic file under `~/.claude/agent-logs/`. Use this skill proactively the moment you've just (a) fixed a bug whose root cause was not obvious from the symptom, (b) confirmed a workaround for an external tool/library quirk, or (c) verified behavior that contradicts docs or comments. Also use when the user says any of these exact phrases or their close paraphrases — "log this", "log this learning", "save this learning", "remember this for next time", "add to LEARNINGS" — or types `/agent-logs`. The skill writes to the appropriate file; do not append by hand.

khe618 Bundle

Curate a project's task queue (the factory checkout's queues/<project>.md for factory-managed projects, else a project-local TASKS.md) into an ordered queue of self-contained, agent-ready tasks. Use when the user asks to create, update, prune, prioritize, or manage TASKS.md or the factory queue; asks to turn IDEAS.md/ideas/<project>.md, recent session logs, or git history into actionable tasks; or wants in-progress tasks removed after they have merged into master/main.

amwangfan Bundle

Control the user's current Chrome profile through the local ego-chrome extension and CLI. Use for navigation, forms, dynamic interfaces, extraction, and browser testing that benefit from existing login state. Minimize execution rounds and model input: keep predictable work in one invocation, prefer semantic locators and bounded DOM reads, and use compact snapshots only when the state is genuinely unknown.

khe618 Bundle

Delegate token-heavy, mechanically-verifiable implementation work to an external coding agent (Codex preferred, opencode on quota exhaustion) to conserve the main session's context. Use when the approach is already decided and the work is bulk — many files receiving the same transformation, a large test suite, a migration, a codemod — and success is checkable by tests, a build, or a greppable invariant. Also use when the user says "delegate this", "hand this to Codex", "offload this", or "run this on opencode". Do NOT use for work under ~3 files or ~100 lines, for anything needing mid-flight conversation, for design decisions, or for diagnosis and debugging handoffs — those go to /codex:rescue.

agentic-lineage Skill

Example skill for reviewing resume material in a packaged agent environment.

danthamanvoiss Skill

Route website creation work to the right web brief, wireframe, audit, motion, or prompt-adapter skill.

malik-xae-12 Bundle

Complete global setup, browser automation, and verification guide for Microsoft's official Playwright MCP server (@playwright/mcp) in Google Antigravity and Claude. Use whenever the user wants to configure Playwright MCP globally or in a workspace, run automated browser tests, verify local development servers (http://localhost:3000, http://localhost:5173), test responsive UI across breakpoints (1440x900, 1920x1080, 1024x768, 768x1024, 390x844, 375x812), test scroll-driven/scrollytelling animations, inspect console errors, or troubleshoot MCP connection and browser launch issues.

danthamanvoiss Skill

Translate loose prompts into structured creative briefs with requirements, constraints, and acceptance criteria.

ilovefiniki Bundle

Initialize a new or existing project according to the Modern Multi-Agent Standard. Sets up AGENTS.md, GEMINI.md, CLAUDE.md, ARCHITECTURE.md, README.md, session scratchpads, fast zero-cost security rules, and runtime browser verification.

danthamanvoiss Skill

Convert approved web requirements into reusable UI prompt scaffolds inspired by superdesign prompt patterns.

rick-orphen Skill

Run a single top-tier model as an independent, read-only reviewer and get findings back — no heavy ceremony. Works as a verifier (deliverable + acceptance criteria → verdict) and as an advisor (a question → an opinion with its grounds). Use when: someone asks for the sharpest single read, explicitly without a committee. Do NOT use when: the answer is genuinely contested and you need several models to argue it out. Never fires automatically.

raulalmeidatarazona Skill

Create and validate reusable agent skills with the standard SKILL.md format.

zj-ai-lab Bundle

用使用者自己的 Word 模板生成《委托代理合同》和配套《授权委托书》，支持单件与成套出件。 当用户要求"做委托合同"、"合同和授权一起出"、"新签客户"、"用我的模板出授权委托书"时使用。 支持民商事自然人/单位、刑事被告人/被害人、单/双律师及仅会见等模板类型，具体范围以用户模板为准。 首次使用（或新增文书类型）先走第 0 步初始化：agent 向使用者要本次需要的合同和授权委托书 Word， 动态提炼出要填写的位置、下划线与 tab 布局，生成模板包；没有模板包不能出件。 收费方式条款由 LLM 把口语描述润色成正式条款（含数字大写），并主动做风险审查（代理范围 / 触发条件 deadline / 后续阶段处置 / 风险代理上限 / 婚姻继承与刑事等禁风险代理红线）。

lwyxzm Bundle

Goal-driven UI verification on iOS simulators, Android emulators, and Chromium: Jev (TypeSafe System One) makes the fast per-step judgments — read the current screen, pick the next tap/type/scroll, and decide whether the acceptance criterion is met — while Maestro executes and Maestro MCP stays available for interactive control. Use when asked to verify a feature in the simulator, reproduce or confirm a UI behavior, drive an app without a pre-written flow, check a screen against a criterion, or turn a verification run into a Maestro flow. Handles compound goals: split them into ordered instructions with --steps (or let --plan split them) and run them in one session. Triggers: "验证一下这个功能", "帮我测一下这个页面", "在模拟器里验证", "看看登录页对不对", "多步骤验证", "verify this screen", "does the app show ...", "drive the app to ...", "turn this into a Maestro test".

latticecast Bundle

Rule for creating portable agent skills (SKILL.md). Use when building, structuring, or debugging reusable skills and commands for coding agents.

latticecast Bundle

Start the autonomous multi-agent dev loop — a queen + bees in tmux solving tickets from LatticeCast PM

latticecast Skill

Persistent cluster that delegates goal-scoped swarms to agentic hives; use for multi-initiative autonomous delivery and recovery

mhiqrambg Bundle

Use when evaluating, critiquing, transforming, or generating a Master's Informatics thesis idea, research gap, novelty, research question, experiment, AI/ML/CV/NLP/LLM/Edge-AI method, or when distinguishing R&D research from software implementation using the Frascati Manual 2015.

latticecast Skill

Onboard a new codebase — explore, understand, and generate .tmp/llm*.md docs for quick reference. Use when joining a new project or needing to understand unfamiliar code.

latticecast Skill

Guide for creating portable AGENTS.md project instructions for coding agents. Use when setting up, writing, or organizing repository-level agent guidance.

latticecast Bundle

LatticeCast PM integration — ticket status updates, project setup, pre-flight checks. Internal library used by developing/programming, agent/agentic-hive, and developing/onboarding.

microsoft Bundle

Multi-dimensional review of a PR or feature branch in microsoft/win-dev-skills. Activate on "review my PR / changes / branch", "vet before pushing", "PR review", "is this ready to merge". Fans out parallel sub-agents over skill content, the skill-vs-tool boundary (solution hierarchy), tool correctness, payload/provenance/tests, docs & manifest sync, plus a multi-model cross-check. Reports findings to stdout. Does NOT apply fixes.

abdulyasir100 Skill

Audit installed Clawdbot/MCP skills for security risks using SkillLens. Detects exfiltration, execution, persistence, privilege bypass, prompt injection, and overbroad triggers. Produces risk-scored reports with concrete evidence.

abdulyasir100 Skill

Production-grade data science and ML expertise. Covers statistical modeling, A/B testing, causal inference, time series, ML deployment with monitoring and drift detection, real-time inference optimization, and end-to-end ML pipelines with Python, PyTorch, TensorFlow, and Scikit-learn.

abdulyasir100 Skill

Reduce latency when an app calls the `claude` CLI as a subprocess (Sonnet or any model) — faster AI turns / responses without changing the model. Use when generation "feels slow", a Claude-CLI-backed app lags, someone compares it unfavorably to a local LLM, or you're asked to speed up / optimize per-call LLM latency, time-to-first-token, or throughput. Logic and decision order only — not implementation. Distilled from a real optimization that cut median latency ~30% and killed random multi-second freezes with no quality loss.

caojingzhi1015-bit Bundle

半调拼贴海报专业 Prompt 生成器。三步走——① 分析原图（吸主色/选强调色/读主体类型/判断对比度高低/读拼贴元素）② 按对比度分路线生成 prompt（高对比→锋利 90 年代杂志风 / 低对比→柔和 ins 文艺风）③ 按固定格式输出。共同规则：60-30-10 配色、装饰取自原图、细实线方框 1-3 个圈局部特写、大标题粗体大写英文放左上或右上、小标签放右下、留白充足、网点 45 度斜角。当用户要求 半调海报、半调风格、网点海报、halftone、波普海报、pop art、丝网印刷风、网版印刷、杂志拼贴海报、scrapbook collage、zine、90年代拼贴、复古拼贴、Y2K 海报、孔版印刷人物海报、爱豆周边，或发来照片说"以这张为参考""重新生成一张""也做成这种效果"时触发。默认同时产出结构化 prompt 与生成图。

fauzanazz Bundle

Use when the user asks about building agents, designing tool sets for LLMs, structuring agent workflows, reviewing existing agent architectures, creating action spaces, coordinating subagents, or evaluating agent behavior. Also trigger when you see terms like "agent harness", "tool calling design", "action space", "agent loop", "progressive disclosure", "subagent", "elicitation", or when the user asks how to make an LLM interact with external tools effectively. Also covers production AI service architecture: "agent orchestration", "sub-agent spawning", "swarm", "recursion budget", "RAG design", "document intelligence". Applies to Claude API agents primarily, but principles are model-agnostic.

wjg-ares Skill

多 Agent 协作编排的首次初始化 —— 建 SQLite 账本、装常驻规则、把本会话注册成主 agent。在某个项目里第一次使用这套编排前必须先跑一次；可重复执行。

wjg-ares Skill

主 agent 的调度台 —— 查看各 worker 在干什么、谁待审批、队列还剩什么，以及建任务、派活、查交付、回收死掉的 worker。compact 之后调一次即可完全恢复调度状态。

golfgolfgolf200 Bundle

Warppay Mcp Skill