Agent Workflow Claude Skills
4 121 agent workflow skills from 665 authors. They configure the agents themselves: memory, prompts, context and other skills. Half of them fit into 1 845 tokens or less — that is what one costs your context window when the agent loads it. 778 ship runnable scripts rather than instructions alone. 5 of them cannot work without an MCP server, most often task. We also found 541 copies of these same skills sitting in other people's repositories — counted once here, not 541 times.
4 121 unique 665 authors 2 767 updated this month 506 from vendors
cheat-on-skill 的核心。连 BOSS 直聘真实招聘数据 + 网页信号，按用户能力画像找「高薪 × 你学得动 × AI 增强」交集里的候选岗位。每个岗位给：薪资量级 / 需求热度 / 你的差距 / 可学性分 / 诚实学习周期，并过 AI 影响分类与反诈红线。触发词："帮我找岗位"/"找AI时代高薪工作"/"有什么我能学的高薪岗"/"skill scan"/"扫一遍招聘"。前置：需要 .skill-state.json（无则先路由到 skill-init）。
cheat-on-skill 的陪跑进度 skill。用户问“今天该干嘛”“我现在做到哪了”“继续学”“打卡”“我卡住了”“下一步是什么”时触发。读取 .skill-state.json 的 active.progress 和 learning_plan，给当天任务、检查完成情况、记录进度、根据快慢调整计划。前置：已有 .skill-state.json 且 active.learning_plan 存在。
> Publishes a local Markdown file to a user-specified Confluence location, creating a new page or updating an existing one through the Atlassian MCP server. Use when the user wants to post, publish, push, or sync a Markdown file to a Confluence space or page. Requires a configured Atlassian MCP server. Does not write or generate the Markdown itself — point it at an existing file, or use project-documentation-to-confluence for the document-then-publish flow, or plan-a-feature-to-confluence for the plan-then-publish flow. Does not publish to Jira — use work-items-to-jira.
> Creates or updates project documentation for a feature, system, or component and publishes it to a user-specified Confluence location. Use when the user wants feature or system documentation written to Confluence, posted to a Confluence space or page, or synced to a Confluence location. Requires a configured Atlassian MCP server. Does not document to local files only — use project-documentation for that. Does not publish an arbitrary existing markdown file — use markdown-to-confluence for that. Does not plan or specify a new feature to Confluence — use plan-a-feature-to-confluence for that. Does not create architectural decision records — use architectural-decision-record. Does not create coding standards — use coding-standard. Does not produce runbooks — use runbook.
> Runs an evidence-based investigation of a bug, failure, or unexpected behavior with investigate and publishes the resulting investigation report to a user-specified Confluence location. Use when the user wants something debugged, diagnosed, or root-caused AND the findings posted to a Confluence space or page. Requires a configured Atlassian MCP server. Does not investigate to a local file only — use investigate. Does not publish an arbitrary existing markdown file — use markdown-to-confluence. Does not document an already-understood feature to Confluence — use project-documentation-to-confluence. Does not plan or specify a new feature to Confluence — use plan-a-feature-to-confluence. Does not publish to Jira — use work-items-to-jira.
> Produces a progressive-disclosure overview of unfamiliar code or a pull request's changes with code-overview and publishes the resulting overview to a user-specified Confluence location. Use when the user wants code or a PR explained, oriented, or made sense of AND the overview posted to a Confluence space or page. Requires a configured Atlassian MCP server. Does not produce the overview to a local file only — use code-overview. Does not publish an arbitrary existing markdown file — use markdown-to-confluence. Does not document an already-understood feature to Confluence — use project-documentation-to-confluence. Does not root-cause a bug to Confluence — use investigate-to-confluence. Does not plan or specify a new feature to Confluence — use plan-a-feature-to-confluence. Does not publish to Jira — use work-items-to-jira.
> Capture structured feedback on the Han skills and agents used in the current session and optionally post it as a GitHub issue to testdouble/han. Use at the end of any session where one or more han-* skills or agents ran, to rate a run, log what worked and what didn't, or submit observations for maintainers. Does not review code, investigate bugs, or research options; use code-review, investigate, or research for those. Does not provide feedback on skills or agents from non-Han plugins.
> Builds a new Claude Code agent (subagent) from scratch through a relentless, evidence-based interview that walks the agent's design tree decision-by-decision — entity fit, domain focus and vocabulary, role identity, anti-patterns, description, model tier, tools, and self-containment — then reviews the finished agent against the plugin-building guidance and applies every fix it finds. Use when creating, authoring, scaffolding, designing, or drafting a new agent or subagent. Does not build a skill or slash command — use skill-builder. Does not serve, vendor, or refresh the authoring guidance itself — use guidance.
> Authoritative guidance for building Claude Code skills, agents, and plugins, plus init and update steps that install and refresh the plugin-building skills in the current repository. Use when you need the rules or best practices for a skill, agent, hook, or plugin — designing, reviewing, hardening, or checking one against the guidance. Run with `init` to vendor the guidance, skill-builder, and agent-builder skills into the current repository (so they run with no dependency on this plugin) plus a path-scoped rule index, or `update` to refresh an already-vendored copy. Does not run an interview to build a new skill or agent from scratch — use skill-builder or agent-builder. Does not write feature code, review application code, or build non-plugin features.
> Builds a new Claude Code skill from scratch through a relentless, evidence-based interview that walks the skill's design tree decision-by-decision — entity fit, use cases, name, description, workflow steps, tools, and progressive-disclosure layout — then reviews the finished skill against the plugin-building guidance and applies every fix it finds. Use when creating, authoring, scaffolding, designing, or drafting a new skill or slash command. Does not build an agent or subagent — use agent-builder. Does not serve, vendor, or refresh the authoring guidance itself — use guidance.
> Authoritative guidance for building Claude Code skills, agents, and plugins, vendored into this repository. Use when you need the rules or best practices for a skill, agent, hook, or plugin — designing, reviewing, hardening, or checking one against the guidance. Does not run an interview to build a new skill or agent from scratch — use skill-builder or agent-builder. Does not write feature code, review application code, or build non-plugin features.
Guide for creating effective skills that extend agent capabilities with specialized knowledge, workflows, or tool integrations. Use this skill when the user asks to: (1) create a new skill, (2) make a skill, (3) build a skill, (4) set up a skill, (5) initialize a skill, (6) scaffold a skill, (7) update or modify an existing skill, (8) validate a skill, (9) learn about skill structure, (10) understand how skills work, or (11) get guidance on skill design patterns. Trigger on phrases like \"create a skill\", \"new skill\", \"make a skill\", \"skill for X\", \"how do I create a skill\", or \"help me build a skill\".
Project-context awareness — multi-session workflows around one topic, shared materials and preferences, and cross-mode handoffs as a high-value user path.
Author a new Pneuma mode end-to-end — manifest + viewer + skill + seed + showcase. Use this skill whenever the user says they want to create a new mode, fork an existing one for a different domain, scaffold mode files, design a new viewer, or asks "how should I build a mode for X". Walks the user through a discovery interview, produces a design brief that names every key choice (Source kind, ViewerAddress vocabulary, action space, seed strategy, external integrations, evolution directive), and only then generates the directory skeleton. Encodes the practice rules pulled from webcraft / slide / diagram / illustrate / remotion / kami. Pneuma Skills project only; Claude Code only.
> Rewrite the active Pneuma session's UI title + one-line summary so the launcher and ProjectPanel rows reflect what the session is actually about. Use this skill whenever the user asks to "整理 / 概括 / refresh / re-title / summarize this session", whenever the conversation has produced substantive work and the default title ("WebCraft session") is now uninformative, or before the user pauses a long session — the next time they reopen it, the row needs to say reopens, mode-agnostic.
> creating modes, editing manifest.ts, pneuma-mode.ts, viewer components, skill prompts, seed files, publishing, forking, or any mode package development. This skill defines the ModeManifest reference, ViewerContract patterns, and publishing workflow. Consult before your first edit in a new conversation.
Use when mining coding-agent session history, archived transcripts, memories, or repeated local work to discover recurring workflows that should become new Agent Skills.
Use when turning local, private, or personal Agent Skills into publishable skills for GitHub, marketplaces, teams, or public sharing, especially when private paths, personal habits, credentials, internal hosts, or user-specific context must be removed.
Use when auditing or adapting newly created, downloaded, forked, installed, or community Agent Skills to the user's tools, habits, directories, session history, and preferred workflows, especially when triggers feel wrong, noisy, or too generic.
>- 质检 Agent：检查 vibe-coding-kit 其他 Skill 的产出物是否符合治理规范。 当用户说"检查一下产出"、"帮我看看合不合规"、"质检"、"验证 PRD"、"检查项目说明书"， 或在任何 Skill 完成产出后，使用此 Skill 做合规检查。 它读取项目根目录的 harness.json，逐条校验产出文件，报告违规项和修复建议。 这是 vibe-coding-kit 套件的治理层（harness），确保所有产出物格式统一、质量达标。 它不止检查格式，还会做跨 Skill 一致性校验——比如 PRD 里的项目名和项目说明书里的项目名是不是一致。
>- 一个"访谈式"的需求 Agent：通过一问一答 + 在岔路口给建议，把一个说不清需求的人，带到一份 AI 能直接照着开工的 PRD。 当用户说"我想做个 XX 但不知道怎么跟 AI 说"、"帮我理需求"、"帮我写个 PRD"、"我有个想法，帮我想清楚再开工"， 或在 Claude Code / Codex 里准备开一个新项目、却还没把要做什么定清楚时，使用此 Skill。 它不是让用户填表格，而是像产品顾问一样访谈：一次只问一件事，遇到用户判断不了的选择（尤其"产品形态"和"技术栈"） 会给出 2-3 个大白话选项 + 推荐一个最简可行的默认 + 一句话讲清代价，等用户拍板。 访谈结束后，直接把 prd.md 和 项目说明书.md 写进项目目录，并提议生成可点击的 HTML 原型。 这是 vibe-coding-kit 套件里"把需求从人嘴里问出来"的交互入口。
Update agent context files for this project. Use when the user invokes /update-agent-context, asks to "update agent docs", "update CLAUDE.md", "update AGENTS.md", "add to agent_docs", "record a correction", or mentions updating context for a specific topic.
Set up Claude Code for this project. Use when the user invokes /setup-claude, asks to "set up claude", or "create the skills symlink".
Build the curated interactive SET the Editor approved: the ONE narrative-bound hero centerpiece (an explorable/scrollytelling that makes the reader PRODUCE the lead finding) PLUS the ranked supporting playgrounds — each bound to a distinct finding — to the same three-layer-number standard, plus animation/transition craft. Consumes editor.json.interactives + imagineer.json + the Analyst's client_model. Outputs interaction.json (centerpiece + supporting[] + playtest_handoff) for the Programmer; may reach back into the Editor's spine when the lead finding should be hands-on.
Use when creating a full or partial set of genre wrapper skills under a genre directory. Builds `.github/skills/` subfolders and `SKILL.md` skeletons for a specified genre, keeps the `题材名-能力名` naming rule, and writes hard requirements that each genre skill must load and use its corresponding common skill when that common skill already exists.
The original v1 taste-skill, preserved for projects depending on its exact behavior. The current default is `design-taste-frontend` (v2 experimental), which is a substantial rewrite. Use this v1 install name only if you need exact backward compatibility.
用于以默认极严、保守、负面证据优先的口径评估作品在目标平台的签约或过稿潜力。适合统一承接评估流程骨架、分阶段准入评估、材料完整度判断、评分维度、概率区间、封顶与一票否决规则，并兼容竞争位 / 竞品威胁评估语境。评估{目标平台}项目时，强制要求使用多种可用网络搜索工具检索{目标平台}同题材 TopN 候选池（起点默认按起点中文网榜单检索，其他平台按可用性调整），并按 "TopN 候选池 + 3–5 主压制样本 + 四层竞对拆解"做从严对照。目标平台路由优先级：用户显式指定 > Agents.md 主输出平台 > 默认回退起点中文网。关键词：平台签约评估、极严评估、分阶段准入评估、总纲预评估、分卷预评估、正文准入、过稿潜力、签约概率评估、概率区间、一票否决、竞品威胁评估、评估报告、榜单对照。"
用于蒸馏一位小说作者的文笔 DNA，产出可被下游写/审/改 Skill 消费的"作者风格模板"。适合把作者的多部作品的文笔特征提炼为可测量的风格约束（句长基线、段落节奏、感官偏好、对话标签偏好、比喻指纹、禁忌清单），并在目标项目根目录自动注册到 Agents.md。关键词：蒸馏作者文风、作者风格模板、文笔DNA、风格注入、文风复刻。
> Security audit and vulnerability scanner for AI agent skills before installation. directory or git repo URL for malicious code, (3) pre-install security gate for Claude Code plugins, OpenClaw skills, or Codex skills, (4) scanning Python scripts for dangerous patterns like os.system, eval, subprocess, network exfiltration, (5) detecting prompt injection in SKILL.md files, (6) checking dependency supply chain risks, (7) verifying file system access stays within skill boundaries. "check skill before install", "skill security check", "skill vulnerability scan".
Skills for web search and content scraping via DuckDuckGo MCP Server. Used when users need online searching and web scraping.
The skill for control Home Assistant smart home devices and query states using MCP protocol.
A test skill
An example skill for testing diff-skill evaluation.
An example skill for testing diff-skill evaluation.
| Compare two versions of a skill's instructions to detect degradations and complexity increases. Asymmetric check — only regressions count. Additions, clarifications, and strengthenings are not degradations. Compares HEAD → working tree by default. Walks the full reachable file set from SKILL.md via markdown links. Produces a report file under docs/reviews/diff-skill/ and an inline summary.
An example skill for testing diff-skill evaluation.
An example skill for testing diff-skill evaluation.
> Use when the user has a prompt that feeds a system they can already score, and wants that prompt automatically improved to raise the score against their own evaluation command. Makes one targeted quality edit per iteration — clarity, context, specificity, structure, examples, decomposition, guardrails — re-runs the user's eval to measure the metric, and keeps the edit only if the metric improves, else reverts; loops to a target, plateau, or budget. The metric is whatever the user's eval command prints (task accuracy, an LLM-judge score, a pass rate, a tool-call success rate); the loop is metric-agnostic and never edits the eval. Not for writing a prompt from scratch, not for tuning model weights or hyperparameters, and not for one-off manual prompt edits without a score.
> Use when the user wants to adversarially stress-test a guardrail, classifier, prompt, or API they own or are authorized to test, to surface the distinct ways it fails. Generates adversarial inputs, runs them through the target and a ground-truth oracle, logs every target-vs-oracle disagreement as a failure de-duplicated by technique class, and loops until rounds stop surfacing new classes. Produces a catalogue of distinct, reproducible failures — the attacker half of a find→fix setup. Not for patching the target, and not for attacking systems the user does not own or have permission to test.
> Use when the user has a coding prompt — a feature, bug fix, or refactor — and wants it implemented end to end by a self-checking software loop, not a single pass. It refines the prompt into an subagent implements each PR-sized task editing only source, a separate QA subagent authors the tests that prove the task's acceptance criteria (or grades quality only when tests already cover them), runs the task's tests plus the full accumulated regression suite, and grades the code against a strict conciseness/readability/style-match rubric; the Engineer and QA loop on each task until the tests pass and the quality gate holds, then it commits and opens a pull request for the task — stacked PRs (one per task, each based on the prior) by default, or one combined PR — and moves to the next task. Author and critic have disjoint write scopes (Engineer owns source, QA owns tests) so neither can game the gate. Not for producing a plan without building it (that is plan-loop), and not for tuning a metric on an existing artifact under a fixed correctness bound (that is optimize-loop).
> Use when the user wants an autonomous ML research loop that pressure-tests competing ideas before spending compute — several research subagents each propose one architecture change, a self-calibrating Judge critiques them against a rubric, the proposers refine, and the Judge picks the single change to run. The Judge learns to pick better over time by scoring its own predictions against realized metric deltas, recording predicted-vs-realized in a calibration ledger and refining its working rubric. The result is an experiment ledger where each iteration's change won a de-biased tournament. Not for running a single pre-decided experiment, and not for analysis-only exploration — for one hypothesis proposed and run per iteration without competition, use the sibling ml-autoresearch loop.
Analyze LLM experiment results. Handles single or comparative experiments, exploratory or Q&A modes. Use when user says "analyze experiment", "compare experiments", "analyze against baseline", or provides one or two experiment IDs for analysis.
Datadog skills for AI agents. Essential monitoring, logging, tracing and observability.
End-to-end Agent Observability pipeline for an instrumented ml_app — classify production traces, root-cause failures, bootstrap evaluators, then (optionally) sample + publish a dataset, generate + run an experiment, and analyze results. Six narrated phases with a standardized banner and a "continue" checkpoint between each. Pure orchestration over the agent-observability sub-skills (`agent-observability-session-classify`, `agent-observability-trace-rca`, `agent-observability-eval-bootstrap`, `agent-observability-experiment-py-bootstrap`, `agent-observability-experiment-analyzer`). Use when user says "run the eval pipeline", "go from traces to evals", "bootstrap evals end to end", "classify then RCA then bootstrap", "build an eval set from scratch", "onboard me to datasets and experiments", "walk me through experiments", "I have an ml_app, now what", "Agent Observability onboarding", "guided experiment setup", "from traces to experiments", or wants a deterministic, narrated tour from production data through evaluators, datasets, and experiments. Stop early with `--stop-after <phase>` to short-circuit at evaluators or dataset, or resume mid-flow with `--start-at <phase>`.
> Classify whether user intent was satisfied in a Datadog Agent Observability trace or session. (2) trace_id — classify a single Agent Observability trace without RUM; (3) ml_app — sample and classify multiple sessions or traces from a given LLM app. Output is compact by default (verdict + one-sentence reason). Use when evaluating satisfaction, classifying sessions/traces, labeling data, or generating signal for agent-observability-eval-pipeline or agent-observability-trace-rca.
>- Run an iterative code-improvement hill-climb against real Datadog LLM-Obs data, locally, with Claude Code as the agent. Establishes a baseline eval, makes one focused change, re-scores with the same harness, keeps the change if it improves the score in the goal's direction (labeling within-noise gains tentative), and repeats. Use when the user says "run an auto experiment", "hill-climb this code", "iteratively improve X and measure the delta", "optimize this prompt/file against my traces", "auto-optimize against LLM-Obs", or wants the local equivalent of the auto_experiments worker. Works from a local dataset file, an ml_app, a dataset_id, or a list of trace_ids.
Generate a live Single Step Instrumentation (SSI) onboarding confirmation report — verifies APM instrumentation is working end-to-end with deep links into the Datadog UI. Only use after agent-install and enable-ssi have both completed successfully.
Root cause analysis on production LLM traces. Diagnoses why an LLM application is failing — works from eval judge verdicts, runtime errors, or structural anomalies depending on what signals are present. Walks the span tree from symptom to root cause. Use when user says "what's wrong with my app", "why is my eval failing", "analyze errors", "root cause analysis", "diagnose failures", or wants to understand production failure patterns.