GITHUB AGENT SKILLS / 技能收录
给你的 AI 助手
装上专业技能
我们从 GitHub 持续收录、审核并分类优质的 Agent Skills，按职业与使用场景整理，复制一行命令即可安装到 Claude Code、Cursor 等智能体。
按职业浏览 Agent Skills
为不同岗位挑选的高频技能
按使用场景浏览 Agent Skills
从你要完成的任务出发
按创作者浏览 Agent Skills
按 GitHub Stars 排序的优秀创作者
Microsoft
@microsoft
openclaw
@openclaw
OpenAI
@openai
The Apache Software Foundation
@apache
GitHub
@github
Anthropic
@anthropics
AutoGPT
@Significant-Gravitas
Garry Tan
@garrytan全部 Skills 192263
source-synthesis
Deep synthesis of a curated reading list — blog posts, tutorials, docs, articles — tailored to a specific use case or build goal. Use whenever the user provides URLs (public links to educational content) and wants a thorough, opinionated synthesis rather than a raw summary. Trigger on phrases like "synthesise these sources", "give me a lit review of", "I want to read up on X, here are some links", "help me digest these articles", "summarise these tutorials for my use case", "I'm about to build X, here's my reading list", or whenever the user pastes a list of URLs alongside a goal or project description. Also trigger proactively when the user provides multiple URLs and a description of something they're trying to learn or build — even if they don't say "synthesise".
prototype
Build a throwaway prototype to answer a design question. Use when the user wants to sanity-check whether a state model or logic feels right, or explore what a UI should look like.
handoff
Compact the current conversation into a handoff document that another agent picks up in a fresh context window — including an agent targeting a different repo. Use when the user says "hand off", "write a handoff", "pass this to another agent", "continue in another repo", or wants to preserve context before switching sessions/repos.
refactor-django
任意 Django 專案的波浪式重構工作流程。當使用者想要讓專案更好維護、程式碼更具可讀性、更容易測試、更 clean、提升測試覆蓋率、從 Fat View 抽取 Service 層、拆解 Fat Model、解耦 Django signals、修復技術債、整備 type hints / mypy、設定 pre-commit 工具鏈、或需要一套安全且可追蹤的重構計畫時使用。觸發詞：refactor、重構、service 層抽取、測試覆蓋率提升、fat view、fat model、技術債清理、clean code、service layer、安全網測試、Django 重構、type hints。
grilling
Interrogates a plan, design or decision one question at a time until nothing is silently assumed. Use when the user wants their thinking stress-tested rather than executed: "grill me", "grill me on this", "poke holes in this", "stress-test this plan", "challenge this design", "what am I missing here", "argue against this", "pressure-test this decision", or "interview me about this before we build". Trigger when the user presents a plan and asks for pushback instead of implementation, even without the word "grill". For a session that also checks the plan against CONTEXT.md and existing ADRs, use grill-me-with-docs instead.
grill-me
A relentless interview to sharpen a plan or design.
grill-me-with-docs
Grilling session that challenges your plan against the existing domain model, sharpens terminology, and updates documentation (CONTEXT.md, ADRs) inline as decisions crystallise. Use when user wants to stress-test a plan against their project's language and documented decisions.
flow-friction-audit
Drives a live web flow through a real browser, turns the run into evidence-backed friction findings, then routes them into the impeccable skill for critique and redesign. Use whenever the user wants to know where a real user gets stuck, confused or slowed down in a running flow — "audit the signup flow", "where's the friction in checkout", "walk through onboarding and tell me what's clunky", "usability audit of the live app", "why do people drop off here". Trigger on "usability", "UX audit", "friction", "drop-off" or "walk the flow" without this skill being named, and when the user only supplies a URL and asks what's wrong with the experience. Trigger before any URL or credentials exist — settling the goal, flow and environment is this skill's own first step, so ask for what's missing from inside the skill. Also use to run the design half alone from an agreed problems doc.
feedback
Turns the user's unstructured feedback about how a product or system behaves into verified, agent-ready GitHub issues. Splits a complaint into points, checks each against the code and recorded intent (ADRs, CONTEXT.md, git history), plays back what it found until the user agrees, then files one ticket per agreed finding under a shared label for /implement or /fleet dispatch. Use when the user says "/feedback", "I don't like how X works", "this isn't usable because…", "X feels wrong", "this behaviour is annoying", "push back on how Y behaves", or otherwise complains about the repo's behaviour or UX without a concrete bug report or spec. Also takes a jam.dev screen recording with the points as bullets. Not for feedback on a skill (use skill-feedback-collector), a single well-understood bug (triage-issue) or a ready-to-file follow-up (file-issue).
brainstorm
Runs a brainstorming session on any topic: a problem to solve, an opportunity, names, an event, a strategy. Generates far more and better ideas than the user would alone by rotating creative techniques, then optionally narrows them. Use when the user says "brainstorm", "ideas for", "ways to solve", "help me think of", "I'm stuck on", "what are my options for", or "let's ideate", even without the word brainstorm. Not for stress-testing one existing plan (use grilling), deciding what to build next in the current repo (use ideate), or turning a chosen idea into a spec.
unslop
Strip AI tells from prose and put voice back in. TRIGGER: load BEFORE writing user-facing prose, not after. Covers a PR title or body, README, docs page, commit message, issue or ticket text, Slack or email message, article, or release note. Trigger on plain requests like "raise a PR", "write this up", "document this", "draft a message". The user will not say "unslop", the prose itself is the trigger. SKIP for code, config, tests, and log output.
adhd
Shape output for a reader with ADHD: lead with the next action, number multi-step work, restate state across turns, suppress tangents, give specific time estimates, make wins visible. Invoke with /adhd; stays on until "stop adhd mode".
triage-reflections
Routes pending REFLECT-marker learnings from the reflect ledger to GitHub issues. Use when the user says "triage reflections", "/triage-reflections", "process my REFLECT markers", "what have I learned", or when the statusline shows reflect:N. Infers missing skill and scope, splits notes into explicit and unclear, plays unclear ones back for correction, merges duplicates into one item with every example, drops lessons already covered by a rule or lint config, routes by a three-way test the user can correct, files one issue per skill, and writes every decision back to the ledger with `reflect marker mark`.
subagent-creator
Designs a lean Claude Code subagent, or widens one on request. Use when the user wants to create, scope or audit a subagent or custom agent, or asks to give an existing one more tools, turns or access. Grills on whether it earns its place before writing anything. Not for writing skills.
skill-updater
Update an existing skill with new context, patterns, or improvements. Edits the skill at its source repo (not the installed copy) and opens a draft PR for the change. Use when user wants to update a skill, modify a skill, add a pattern to a skill, improve an existing skill, edit a SKILL.md, tweak skill instructions, change skill behaviour, or mentions "update skill", "edit skill", "change skill", "add to skill". Also use when invoked programmatically with a skill name, path, and improvement_suggestions[] payload from skill-feedback-collector, with a GitHub issue number, or when /implement hands over an issue whose changes are all under `skills/`. Do NOT use for creating new skills from scratch — use skill-creator for that.
skill-feedback-collector
Collect structured feedback on a skill session and convert it into improvement proposals, then file them as reflect-format issues or hand them to skill-updater. Use when a skill finishes and asks whether you want to log feedback, or when the user says "log feedback on the X skill", "record feedback on skill Y", "capture feedback for skill Z", "give feedback on the X skill", or "I want to leave feedback on skill X". Make sure to use this skill whenever feedback or impressions on a skill are mentioned — even if the user doesn't explicitly say "feedback".
openspec-verify-change
Verify implementation matches change artifacts. Use when the user wants to validate that implementation is complete, correct, and coherent before archiving.
skill-brief
Turns a rough idea for a new skill into a skill-creator-ready brief — the prep work a careful person does by hand before building: source prior art, decide what's worth lifting vs. what only half-fits, check the idea doesn't duplicate a skill you already have, and package the decisions. It does NOT write the final skill; it produces the brief skill-creator builds from. Use when the user says "skill-brief", "brief for a skill", "source and synthesise skills for X", "what's out there before I build a skill for X", "I have an idea for a skill, help me scope it", "capture this into a skill brief", or "turn this into a skill-creator prompt". Trigger proactively whenever someone is about to start a skill-creator session from a raw idea and hasn't sourced prior art yet — the sourcing pass is exactly what prevents building a duplicate.