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
@microsoftopenclaw
@openclawAnthropic
@anthropicsOpenAI
@openaiThe Apache Software Foundation
@apacheGitHub
@githubAutoGPT
@Significant-GravitasGarry Tan
@garrytan全部 Skills 482556
ci-triage
Find and diagnose RED CI with the gh CLI — answer "is anything failing?" and "why?" correctly, without being handed a URL. Invoke (via the Skill tool) whenever you need to check whether a branch, commit, or PR is green, whenever someone reports failing CI, before cutting a release, and before claiming any commit is green. THE FAILURE THIS SKILL EXISTS TO PREVENT: `gh run list` reports a RUN-level `conclusion`, and a run whose rollup is `cancelled` can contain a job whose conclusion is `failure` — so filtering runs on `conclusion=="failure"` silently misses real red, and taking the latest run per workflow hides an older failed run on the same SHA. The authoritative instrument is the commit's CHECK-RUNS, which is what the GitHub UI renders. Pairs with `ci-watch` (blocking until a run is terminal) and `ci-adhoc-test` (running a probe on a real OS).
ci-adhoc-test
Run ad-hoc / exploratory tests on a real OS or platform via CI when the behavior CANNOT be reproduced on the local host or in Docker — macOS Seatbelt / sandbox-exec / codesigning, Windows cmd.exe / --script-shell shell selection / .cmd resolution / Authenticode, musl-vs-glibc, Linux-arm64, a specific Node floor. Invoke (via the Skill tool) whenever the user asks to "test this on macOS/Windows CI", "probe this across operating systems / platforms", "run an ad-hoc cross-platform check", or any validation that needs a real macOS or Windows runner. THE KEY FACT this skill exists to carry: a pull request is NOT required — a branch-scoped GitHub Actions workflow (workflow_dispatch + push to the branch) runs the probe with no PR open. Pairs with `ad-hoc-test` (local host probing), `dev-loop` (build), `ci-watch` (await the run), and AGENTS.md's Docker section (Linux-only; this skill covers what Docker can't).
catalog-probe
暂无简介
benchmarking
暂无简介
babysit
暂无简介
audit-thread
Use when running a compatibility/parity AUDIT — enumerating where nub diverges from a reference it claims parity with (pnpm CLI grammar, a lockfile format, a Node behavior, a flag surface). Encodes the hard gates that stop an audit from surfacing false positives. Auto-triggers on "audit", "compat audit", "parity audit", "find all the gaps", "what are we missing vs <tool>".
aube-bump
Bump nub's vendored aube engine to a newer jdx/aube upstream. Invoke (via the Skill tool) whenever you need to pull upstream aube changes into `vendor/aube/**` — a new release, or arbitrary commits. Encodes the venue-less merge (build an ephemeral merge commit from nub's own objects with `git commit-tree`; no clone, no `nub-fork`, no push-back) and the two-conflict-list methodology: git's markers find TEXTUAL conflicts, the compiler finds SEMANTIC ones, both lists are machine-generated and exhaustive, so you never hand-review a diff. Also covers merge-base correctness (the single biggest lever), merge-not-rebase, `rerere`, the nub delta inventory that must survive, and the nub-side breaks upstream feature-gating causes. Supersedes the older `aube-sync` skill.
address-issue
End-to-end playbook for working a GitHub issue (or bug-fix PR) on nubjs/nub: triage → acknowledge an external report with an "Investigating" comment → reproduce it yourself → SIZE it → fix it at that size (most issues are a small fix you just write — no investigation spike, no sub-agent fleet, no top-tier model) → open a PR that references the issue with `Closes #N` → on merge, comment the resolution → on release, comment the version + release link. Invoke (via the Skill tool) whenever you pick up an issue to work, or are asked to fix a reported bug. Encodes both the maintainer-hygiene conventions in AGENTS.md (so the reporter is acknowledged, the issue auto-closes on merge, and the loop closes when the fix ships) and the proportionality gate that keeps an ordinary bug from being worked like an audit.
ad-hoc-test
Verify new nub functionality end-to-end by building the dev binary and exercising it against real throwaway fixtures. Invoke (via the Skill tool) in BOTH directions, and the second is the one that gets skipped. (1) CONFIRM — after implementing or changing a subcommand/flag/behavior, check the feature ACTUALLY works, not just that tests pass. (2) FALSIFY — before opening a PR on behavior, and again before calling a review round done, SELF-REVIEW by sweeping many adversarial fixtures for what the change BROKE somewhere you were not looking: boundary abuse, real installed registry packages, monorepo and symlink layouts, the tier x module-format matrix. Green gates plus a working reported-case is NOT that sweep. Reviewers read code and cannot run it, so they systematically miss the silent wrong answer — the resolution that returns a different module with no error — which only a fixture run can catch, by checking WHICH file answered rather than that something did. The loop: create fixtures in a tmp dir, build the dev `nub`, run against them, and diff every result three ways — plain node, a build of your branch's MERGE-BASE, and your build. Use the merge-base, NEVER the shipped release: the release can trail `main` by dozens of unrelated commits, so a difference against it is not attributable to your change and every "pre-existing, not mine" verdict resting on it is unfounded. Ad-hoc e2e is a valid verification method on its own; this skill also covers when to promote a durable check into the committed test suite. Pairs with the `dev-loop` build skill and AGENTS.md's pre-push loop.
autopilot-issue-accept
Use this skill to accept ONE microsoft/apm issue with a declared scope. The canvas Accept button, or a maintainer asking to accept an issue, starts it. Draft only from the latest triage proposed scope, get exact-text approval in this chat from a GOVERNANCE roster human, post that comment as that human, then add status/accepted. Do not invent scope, do not accept pull requests, and do not implement the issue.
autopilot-issue-accept
Use this skill to accept ONE microsoft/apm issue with a declared scope. The canvas Accept button, or a maintainer asking to accept an issue, starts it. Draft only from the latest triage proposed scope, get exact-text approval in this chat from a GOVERNANCE roster human, post that comment as that human, then add status/accepted. Do not invent scope, do not accept pull requests, and do not implement the issue.
pr-description-skill
Use this skill to write the PR description (PR body) for any pull request opened against microsoft/apm. Produces one self-sufficient GitHub-Flavored Markdown artifact: TL;DR, Problem (WHY), Approach (WHAT), Implementation (HOW), validated mermaid diagrams when structure is non-trivial, explicit trade-offs, validation evidence, and a How-to-test section -- with every WHY-claim backed by a verbatim quote from PROSE or Agent Skills. Activate when the user asks to "write a PR description", "draft a PR body", "open a PR", "fill in the PR template", or any equivalent.
autopilot-pr-triage-worker
Use this skill to triage ONE microsoft/apm pull request already selected by autopilot-pr-triage-scheduler. Covers community PRs that fix an issue and PRs opened without an issue. Return one advisory recommendation, never human approval, never a diff review, never merge. Not a queue manager and not autopilot-pr-review-worker.
autopilot-pr-review-worker
Use this skill to run a multi-persona expert advisory review on a labelled pull request in microsoft/apm. panel-mode (full | lean | delta) selects a surface-gated roster plus a CEO synthesizer; omitted or unknown mode is lean. Do not spawn inactive stubs. The orchestrator is the sole writer to the PR: ONE recommendation comment, no verdict labels, no merge gating. The panel is advisory -- it surfaces findings, prioritizes follow-ups, and renders a ship-recommendation that the maintainer and author weigh.
autopilot-issue-triage-worker
Use this skill to triage ONE microsoft/apm issue already selected by autopilot-issue-triage-scheduler. Return one advisory recommendation and a proposed scope brief, never human approval. Do not implement the issue, manage the backlog, or list the queue.
pr-description-skill
Use this skill to write the PR description (PR body) for any pull request opened against microsoft/apm. Produces one self-sufficient GitHub-Flavored Markdown artifact: TL;DR, Problem (WHY), Approach (WHAT), Implementation (HOW), validated mermaid diagrams when structure is non-trivial, explicit trade-offs, validation evidence, and a How-to-test section -- with every WHY-claim backed by a verbatim quote from PROSE or Agent Skills. Activate when the user asks to "write a PR description", "draft a PR body", "open a PR", "fill in the PR template", or any equivalent.
autopilot-pr-triage-worker
Use this skill to triage ONE microsoft/apm pull request already selected by autopilot-pr-triage-scheduler. Covers community PRs that fix an issue and PRs opened without an issue. Return one advisory recommendation, never human approval, never a diff review, never merge. Not a queue manager and not autopilot-pr-review-worker.
autopilot-pr-review-worker
Use this skill to run a multi-persona expert advisory review on a labelled pull request in microsoft/apm. panel-mode (full | lean | delta) selects a surface-gated roster plus a CEO synthesizer; omitted or unknown mode is lean. Do not spawn inactive stubs. The orchestrator is the sole writer to the PR: ONE recommendation comment, no verdict labels, no merge gating. The panel is advisory -- it surfaces findings, prioritizes follow-ups, and renders a ship-recommendation that the maintainer and author weigh.