AI & ML

harishkotra Skill

Use when you need to run a research task on behalf of a user or another agent and return clean results.

harishkotra Skill

Use when you need AI agents to make local government operations more efficient.

alikon-art Skill

创建、查看、更新、暂停、恢复、立即运行、删除或排查 DeterminFlow Cron 自动化任务时必须加载此技能；也适用于 once/interval/cron 调度、时区、Agent 类型与权限、重复次数、静默输出、失败重试和历史输出核验。

alikon-art Skill

查看、创建、修改、删除或排查 DeterminFlow Prompt Template 与 system prompt section 时必须加载此技能；也适用于 section 顺序、workflow_only/chat_only、cache break、自定义 template_variables、系统变量渲染以及 Agent Definition 的 prompt_template 绑定。

alikon-art Skill

创建、修改、删除或排查 DeterminFlow Agent Definition 时必须加载此技能；也适用于选择 agent_type、最小工具权限、Prompt 模板绑定、模型覆盖、Skill/Rule 可见分组、子会话可用性与 Workflow Agent 节点配置。

bbigger004 Bundle

画图表、数据可视化、生成统计图的唯一入口。任何时候用户想要「画个图」「做个图表」「可视化一下」「出张图」，或需要柱状图/折线图/饼图/散点图/热力图/桑基图/KPI 卡片等，都必须用这个 skill。底层是 Flint 图表 MCP（flint__render_chart 等工具），支持 95 个图表模板、3 个可渲染后端（vegalite/echarts/chartjs）、10 套品牌主题。本 skill 的关键是让生成的图**直接显示在对话里**。Also trigger for 'plot this', 'make a chart', 'visualize this data', 'render a chart'.

ramsnerm Skill

LLM-Wiki Framework

tekno-imaji-nyata Bundle

Run a Fable hub that plans and delegates work to Sonnet and Opus spokes through the /hub command and the hub-spoke Workflow in Claude Code. Use when a request should be scouted, split into file-owned units, executed on the cheapest capable model, and adversarially verified while Fable spends tokens only on planning.

tekno-imaji-nyata Bundle

Brainstorm and debate with Fable in super.engineering, then let Fable route workers between Sonnet, Opus, and Fable by task difficulty while Codex orchestrates, reviews, and verifies. Use when the user requests this Fable-led multi-agent workflow or invokes this skill.

mlsecopshub Skill

Improve the prose, headings, and structure of a README.md file. Use when the user asks to clean up or polish a README.

mlsecopshub Skill

Improve the prose, headings, and structure of a README.md file. Use when the user asks to clean up or polish a README.

moqun188 Bundle

项目上下文中心。记录项目、服务器、任务，实现跨 agent 工作无缝衔接。 触发词：记录项目、查看项目、连接服务器、当前进度、接手工作、更新状态、项目列表、服务器列表、工作上下文、技术雷达、有什么新技术、GitHub趋势。 支持自动进化（会话结束自动捕获变更）和技术雷达（自动追踪 GitHub 热门 AI 项目）。

lucasleduc Bundle

Analyste de prompts expert spécialisé Claude/Anthropic. Audite la qualité d'un prompt avant exécution : notation multicritère, recommandations d'amélioration, adaptation au modèle cible (Claude Code, Claude.ai, API Anthropic), et analyse du risque d'hallucination + solidité logique de l'idée. Utilise ce skill dès qu'un projet conséquent démarre, qu'une tâche complexe est lancée, qu'un prompt long ou stratégique est soumis, ou que l'utilisateur demande un audit de prompt. Ne PAS déclencher sur les questions simples ou les échanges conversationnels courts. Déclenche aussi quand l'utilisateur mentionne : 'analyse ce prompt', 'audit prompt', 'optimise mon prompt', 'est-ce que ce prompt est bon', 'comment améliorer ce prompt', ou quand un prompt dépasse ~100 mots et vise une production complexe (code, contenu, analyse, stratégie). Ce skill est le garde-fou qualité avant toute action majeure.

lucasleduc Bundle

Skill de production créative : transforme un brief stratégique en contenu texte + prompt visuel IA. Produit du copywriting senior et de la direction artistique pour IA génératives (Leonardo, Stitch, Midjourney, DALL-E, Flux...). Utilise ce skill dès que l'utilisateur demande de créer du contenu créatif complet (texte + visuel), de produire un post pour les réseaux sociaux, de générer un prompt visuel expert pour une IA générative, ou de transformer un brief validé en livrable. Si aucun brief stratégique n'existe, appelle d'abord le skill `stratege` pour en produire un. Même si l'utilisateur ne dit pas 'directeur créatif', utilise ce skill quand la demande combine texte et visuel, ou quand un brief créatif doit devenir un livrable concret.

lucasleduc Bundle

Gate de qualité pour le RAG jarvis-rag-local. Audite toute note ou batch de notes avant ingestion pour éviter la pollution du graph (zombies, duplicats, contradictions, bruit). Produit un verdict APPROVE / WARN / REJECT avec diff des entités/relations qui seraient créées. Utilise ce skill dès que l'utilisateur veut ingérer dans le RAG, auditer un dossier de notes avant ingestion, vérifier la qualité d'un ajout récent, ou quand un autre skill (curateur RAG, workflow sessions→RAG, agent veille Telegram) a besoin de valider du contenu. Déclenche aussi sur : 'audite avant ingestion', 'check qualité RAG', 'cette note est-elle prête pour le RAG', 'garbage-in-out', 'quality-guard', 'batch pré-ingestion', 'ingestion zombie', 'ingestion duplicat'. NE PAS utiliser pour : indexation simple, lecture du RAG, query Hermès.

lucasleduc Bundle

Outil d'analyse et de reverse engineering de contenu. Déconstruit n'importe quel contenu (post, page web, email, prompt IA, output de skill) pour en extraire la recette : ton, structure, leviers psycho, techniques d'écriture, direction artistique. Produit un rapport d'audit structuré réutilisable par les skills de production (stratege, directeur-creatif, planificateur-social, motion-reel). Utilise ce skill dès que l'utilisateur veut analyser un contenu existant pour comprendre ce qui fonctionne, auditer la sortie d'un autre skill, reverse-engineer un contenu concurrent, extraire un style à reproduire, ou créer un prompt à partir d'un exemple. Même si l'utilisateur dit simplement 'analyse ce post' ou 'décortique ça' ou 'c'est quoi le secret de ce contenu', utilise ce skill.

lucasleduc Bundle

Expert en prompting pour génération vidéo IA. Produit des prompts optimisés et paramétrages pour Kling (2.5 Turbo / 2.6 / 3.0), Seedance 2.0 (ByteDance), Veo 3/3.1 (Google DeepMind), et Nano Banana (Google/Gemini images). Utilise ce skill dès que l'utilisateur veut créer une vidéo IA, générer ou optimiser un prompt vidéo, produire un storyboard pour animation, ou travailler avec n'importe quel outil de text-to-video / image-to-video. Même si l'utilisateur ne dit pas 'prompt' ou 'vidéo IA', utilise ce skill quand la demande implique : produire du contenu vidéo via IA générative, améliorer la qualité de rendus vidéo, choisir entre outils de génération vidéo, ou ajouter un voiceover à une vidéo IA. Déclenche aussi quand l'utilisateur mentionne : Kling, Seedance, Veo, Nano Banana, text-to-video, image-to-video, prompt vidéo, motion brush, anti-AI look, ou video gen.

a-saven Skill

Orchestrate a manual-QA pass (functional or design) on the running app in ANY project — a WEB app in a browser or a NATIVE iOS app in the Simulator. Detects whether the project is a single app or a monorepo, remembers a dev URL and login credentials PER APP, plus an optional read-only DB, asking only for what isn't saved yet (and remembering "declined" so it never re-asks). Then scopes the run to the app(s) being tested and invokes the manual-qa agent. Use when the user asks to "QA this", "verify the app works", "test the flow", "check if X works / looks right in the browser", "test the native app / in the simulator", or invokes /qa-run. Runs in the MAIN thread (it needs to ask the user questions); it sets up context, then delegates the click-through to the manual-qa subagent. Also supports an unattended task mode where a parent injects a per-task URL/port (or Simulator UDID) + worktree and no questions are asked.

a-saven Skill

Orchestrate non-trivial engineering work by scoping it, decomposing it into a parallel task DAG, delegating each node to the cheapest model tier that can meet its acceptance criteria, and verifying before integration. Use this whenever a task involves more than a single mechanical edit — implementing a feature, a cross-cutting refactor, a bugfix with an unknown root cause, a migration, or anything where you are about to spawn subagents or start editing several files. Also use it when the user says "plan this", "break this down", "delegate", "route this", "use subagents", "orchestrate", or hands over a vague ask like "add auth" that needs pinning down before any code is written. Reads the repo's CLAUDE.md for stack conventions and forwards them into every delegated brief. Written for a top-tier orchestrator model; do not skip it just because the task looks tractable solo — routing is the point.

a-saven Skill

Playwright QA (headless, via Playwright MCP)

a-saven Skill

Check, seed, or repair the CodeGraph index and Graphify graph across a repo's git worktrees. Use when a session is running in a worktree and CodeGraph looks absent or wrong, when the user asks whether the graphs are fresh / "check all worktrees", after creating a worktree (Orca, `git worktree add`, or an agent harness), or when a graph answer disagrees with the code actually on disk. Also covers why a Graphify graph goes stale after `git pull` and how to force a refresh.

hewi333 Bundle

Agent-initiated spend via Stripe Link — requests purchases, issues virtual cards, owner approves in the Link mobile/web app. Hermes cannot self-approve. The 'spend' rail to complement stripe-payments.

yyyyyhhhhh0639 Bundle

Use when 创建/修改技能前。SKILL.md 写作规范：frontmatter、正文形态、术语一致、查重。

docfrawley Bundle

Audits AI and agent exposure across the local agent-harness environment and, where applicable, the codebase it works in. Inventories AI frameworks, agent runtimes, MCP servers, credentials, ambient identities, hooks, skills, and plugins; maps downstream reach; and identifies boundaries that are asserted but not enforced. Use for AI/agent security audits, MCP exposure, credential scoping, agent permissions, egress, blast radius, shadow AI infrastructure, prompt-injection reach, harness security, or questions about what an agent or framework can reach. Not a diff review.

lbachelotcapitalb Bundle

Run a terminal command that needs a secret only the user knows by collecting it through a native hidden-input dialog piped straight into the command — never asking the user to paste it. Use when a command would otherwise block on a credential: deploying or rebuilding an encrypted artifact, decrypting a vault, logging into a CLI, an SSH key passphrase, or a `sudo` step. The secret stays in RAM for one command and is never written to disk or history. Déclenche AUSSI sur le symptôme, sans qu'on te le demande : « coffre verrouillé », « BW_SESSION absent », « mot de passe introuvable », un serveur MCP qui refuse faute d'identifiant, `bw status` = `locked`, `Permission denied (publickey)`. Un secret verrouillé n'est jamais un « non mesuré » : ouvre la fenêtre masquée, seed la clé, rejoue l'appel. GESTE 0 BLOQUANT, avant d'écrire la moindre ligne de commande : `secret-agent.mjs keys` — le secret est peut-être déjà en RAM. Ouvrir une fenêtre sur un secret déjà mémorisé est un défaut, pas une précaution.

haumer Bundle

focus — show where the agent is looking

why20261 Bundle

分析视频内容，支持今日头条、抖音、B站、西瓜视频、小红书、快手等平台。自动下载视频、语音转文字、繁简转换、支持普通话、粤语等方言及外语、支持提取音频文案、提取口播文案、理解场景描述文案。当用户提供视频链接并要求分析视频内容、总结视频、看看视频讲什么时使用此技能。调用大模型自动转写并剔除语气词、口误与重复内容，并可通过自定义 Prompt 生成总结、改写、金句提取、分镜头、中英翻译等风格化文案。

leichu0612-byte Bundle

Inspect, normalize, structurally chunk, enrich, and validate Markdown or text corpora before embedding or vector-store ingestion. Use when preparing RAG datasets, diagnosing chunk quality, or producing traceable chunk JSON and a quality report; do not use it to invent missing source metadata or claim retrieval gains without an evaluation.

why20261 Bundle

分析视频内容，支持今日头条、抖音、B站、西瓜视频、小红书、快手等平台。自动下载视频、语音转文字、繁简转换、支持普通话、粤语等方言及外语、支持提取音频文案、提取口播文案、理解场景描述文案。当用户提供视频链接并要求分析视频内容、总结视频、看看视频讲什么时使用此技能。调用大模型自动转写并剔除语气词、口误与重复内容，并可通过自定义 Prompt 生成总结、改写、金句提取、分镜头、中英翻译等风格化文案。

realzillionx Bundle

Use for Inspire/启智平台 (qz.sii.edu.cn) through the inspire CLI: install/update/uninstall, accounts, local SII proxy setup, INSPIRE.md asset upkeep, workspace/project/resource/path selection, Notebook, GPU Job, HPC, Ray, Serving, TensorBoard, Image, Model Registry, observation, cleanup, and Inspire CLI Browser API maintenance. Use CLI Help for syntax and load only the focused reference.

enrikkk Skill

Write an end-of-session/end-of-task progress record so a future agent starts with full context. Invoked by the user with /log-progress. Logs to BOTH the project memory directory and the project's own state doc. Records what was done, what was decided and why, the numbers that were measured, the traps discovered, and — importantly — what was left as future work.

enrikkk Skill

Run a deep testing pass on a feature (or set of features) after it's implemented — functional correctness, edge cases, data integrity, the project's own automated suite, live/real-environment behavior as an actual user would hit it, AI/agent-specific risks when relevant, and whether the change broke anything else. Invoked by the user with /test-features [what to test], or whenever they ask to "test what we just built," "verify this feature," or run a "testing session." With no argument, the scope is whatever was implemented in the just-finished coding session — inferred from the diff/recent changes and the conversation, cross-checked against each other. If the scope is ambiguous (several distinct changes, unclear which one is meant), ask the user rather than guessing. This is a verification skill, not a bug-fixing or test-authoring skill by default — see the bug-handling and coverage-gap rules below for the exact boundary.

enrikkk Skill

Walk the user through resolving git merge conflicts between two branches interactively — one file at a time, one conflict at a time, each decided by an MCQ with a real recommendation. Invoked by the user with /merge-branches <branch-a> <branch-b>, or whenever they ask to merge, reconcile, or integrate a divergent branch (especially a co-founder's or teammate's) and want to review the actual content rather than accept git's or a model's silent auto-resolution. Handles both ordinary content conflicts and add/add conflicts (both sides independently built the same file from scratch since the branches diverged).

tzwkb Bundle

译员管理系统的自动化 agent——从微信/企业微信抓取译员动态，抽取后写进系统。档期等低风险直接更新；语言对费率、PO、PO 状态等钱相关只提建议并进待审队列。支持固定语言代码自由组合、语言对独立费率、PO 实际字数口径、严格日期/月格式校验。触发词：译员系统自动同步 / 从微信更新译员 / 定时抓译员动态 / translator agent。

gyy489 Bundle

Use the deterministic Two-Headed-Wu capability interface to resolve allowed Skills, tools, workflows, project bindings, managed resources, and the owner's private small notebook. MUST use when the user explicitly says 两头乌/two-head-wu, says “记下来…”, “写到小本本上…”, or “记到小本本上…”, asks to recall/correct/forget the small notebook, asks a project Agent to obtain capabilities from 两头乌, or needs registered knowledge, server, deployment, backup, Siri, or cross-runtime resources managed by 两头乌. Do not use for ordinary project work that needs no Two-Headed-Wu-managed capability.

gyy489 Bundle

Manage the operator's Codex identities through the Two-Headed-Wu agent-identity capability. Use for registering or launching identities, exact cross-account thread continuation, VS Code project-terminal recovery by exact UUID, verified UUID/archive/native-fork metadata, native named-thread resume and terminal-title sync, shared conversation pools, explicit project routing, the fixed owner-only Air/WeChat pool, one-shot or continuous quota-reset sleep insurance for the current task, or unconditional local-time scheduling through OpenClaw cron. Never inspect, copy, export, archive, refresh, or rotate credentials, passwords, tokens, or Keychain entries. Quota and turn-state access is limited to the bounded probes described here; never store or expose turn content.