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 196852
hitl
A human is looking at the work right now and talking while they look. Re-read the tree before answering, treat every observation as a finding to be captured rather than a task to start, and do not touch anything until they say go. Use for /hitl, and whenever the user is walking through recent changes, reviewing a diff, checking a running service, or firing off several reactions in a row — 'what is this', 'why did that change', 'that looks wrong', 'hold on'. Ends by proposing one plan for approval, never by having quietly acted.
hf-cli
Use the Hugging Face Hub CLI (`hf`) for Hub authentication, models, datasets, Spaces, repositories, storage, jobs, endpoints, and related Hugging Face work.
handoff
The precursor to putting work in front of a human: sync the docs to what the code now does, run the project's checks when the change was architectural or a person is about to verify it, and commit. Use before human-in-the-loop verification or a demo, when significant work has landed, or when asked to 'doc sync and commit', 'get this ready to look at', or 'prep for HITL'. Runs across any project — Python, Node, or otherwise — by reading the project's own AGENTS.md for its check commands.
execute-plan
Execute an approved multi-step implementation plan safely, including dependencies, isolated work, review, and final integration.
exec-summary
Produce a standalone executive summary for a named third party who was NOT in the conversation — a stakeholder, a reviewer, a teammate picking the work up, or yourself in three weeks. Gathers its own evidence from the repo and the running system, marks what is measured versus believed, and ends on the decisions the reader owns. Use for /exec-summary, a status report, a handover note, a write-up for someone else to read, or 'where does this project stand'. NOT for compressing what was just said to the person you are talking to — a bare 'bro', 'brief' or 'tldr', or any request to shorten, recap or tl;dr the current conversation, is /brief instead; this skill is chosen only when the reader is someone other than the current user.
dependency-currency
Periodic maintenance pass over a project's dependencies, tooling, pins and agent config: find what is unused, vulnerable, broken or misdescribed, upgrade what is safe, and name what needs a human. Use for 'update the dependencies', 'tooling update', 'are we current', 'check for outdated packages', a scheduled maintenance run, or before a release. Works in any project — Python, Node, or both — by reading the project's own AGENTS.md for its check commands rather than assuming a stack.
dataprep
Run the max_llm data-preparation pipeline (`python -m src.data.preparation`) to tokenize and stage a dataset from a data-prep config, then report the output artifact paths and token counts. Use when asked to prepare, tokenize or stage a dataset; takes an optional config name or path.
create-workflow
Design a deterministic, resumable workflow for multi-step or parallel agent work.
chat
Resolve a trained max_llm checkpoint and its matching config, then print (never run) the command that starts an interactive chat session with it (`python -m src.inference.chat`). Use when asked to chat with, talk to or try out a trained model or run; takes an optional run name such as `p4_norm_ab_rms`.
brief
Re-state the previous assistant message — or, with `session`, the conversation so far — as an executive brief for a technical reader: bottom line first, facts verbatim, evidence over adjectives, open items and required decisions called out. Use /brief, or when asked for a tl;dr, the bottom line, 'where are we', or a summary of what just happened. A message consisting only of `bro`, `/bro`, `brief` or `tldr` IS this request — run it; never reply asking what was meant. This skill is for the person already in the conversation and never runs tools; a summary written for someone who was NOT here — a stakeholder, a handover, a status report to send onward — is /exec-summary instead.
architecture-rules
Check and uphold a project's deterministic architecture rules — who may import whom, and the code shapes that have caused repeat defects. Use before finishing a code change, when adding a module or a cross-package import, and whenever asked to verify the architecture. The project's AGENTS.md names its checks; this skill's project section repeats them.
agent-readable-secrets
Use when agent-readable project content may expose third-party data, credentials, risky source text, prompts, or safety lexicons; when changing read guards; or when reviewing privacy/redaction boundaries.
train
Launch a new max_llm training run on both GPUs (RTX 4090 + RTX 3090 Ti) through torchrun DDP, resolving the TOML config by name or path. Use only when the user explicitly asks to start, launch or begin a run — "resume", "continue" or "keep training" mean the existing run; takes an optional config name or path.
serve-live
Before a person looks at a running service, make sure what is running IS the code in the checkout. Restart the services your change actually touched — not every service — and verify each is serving the current commit rather than merely answering; check the remaining dependencies are up without bouncing them. Use whenever you finish work a person will view in a browser or client, whenever you say a page or service is ready, after any commit that changes served code, and when asked to 'restart', 'refresh the services', 'is it live', or 'serve the latest'. Also covers waiting for lazy warm-up before calling a service ready.
review
Review a local change, branch, pull request, or stack for correctness, regressions, security, tests, and unnecessary complexity. Use when asked to review rather than implement.
regression-coverage
Reshape a slow or brittle test suite around observable user journeys and published contracts. Use when pruning tests, repairing weak smoke coverage, or addressing repeated feature regressions.
pr-babysit
Monitor selected pull requests and report or safely act on review, CI, mergeability, and stack state.
keel
Use a project's local keel code graph to assess change impact, find callers and reuse candidates, validate a plan, or check an edit. Applies only in a repository with a .keel/ directory; skip it for prose-only work.