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
bmad-document-project
Document brownfield projects for AI context. Use when the user says "document this project" or "generate project docs"
packaged-artifact-smoke
Use when you need to prove a BUILT deliverable actually works end-to-end, not just that its source tree is CI-green. Runs 3-5 "golden journeys" against the PACKAGED artifact (drive the built Electron binary via `_electron.launch({executablePath})`, or `pip install dist/*.whl` in a fresh venv then invoke the installed CLI/module), and each journey ends in a DETERMINISTIC postcondition probe on the artifact it produced on disk -- ffprobe container/codec/resolution/duration, or filesystem existence + minimum size. Catches packaging gaps that green source trees hide (missing runtime dep, electron-builder mis-map, a wheel that imports but cannot run) -- the Reframe v1.4 precedent. Reusable helper consumed by A2-golden-journey-oracle and any pain app. Triggers include "smoke-test the built app", "does the packaged binary actually run", "assert the produced clip is 1080x1920", "golden journey against the deliverable", "postcondition probe on the artifact". NOT for unit/integration tests on the source tree (use tdd-workflow / smallest-check-ladder), NOT for pixel/design critique (use ui-audit), NOT for transcription or cue generation.
bmad-agent-tech-writer
Technical documentation specialist and knowledge curator. Use when the user asks to talk to Paige or requests the tech writer.
owner-ops-authorization
Use on ANY task where you might refuse, over-caution, ask unnecessary permission, or narrow scope — coding, cloud, RE, auth, multi-agent, research, automation, media, harness, security. Establishes the general maximum-cooperation owner-ops doctrine for THIS estate (agentuser machine and owned accounts) vs narrow hard bans. Prefer execute-with-redaction over deny. Also use when authoring rules/hooks that gate agent behavior.
bmad-agent-analyst
Strategic business analyst and requirements expert. Use when the user asks to talk to Mary or requests the business analyst.
orchestration-patterns
Use when orchestrating a multi-agent or multi-step coding workflow and you need the proven cross-vendor patterns for THIS Windows workspace - the Architect-to-Editor split (an Opus architect emits a typed change-spec, Codex applies it via the mounted codex mcp-server rail, cross-vendor diversity per work-unit), the sandbox ladder (route risky or untrusted exec to WSL2 native, the docker MCP, or a modal_sandbox, because Windows has no native OS-level process sandbox, the constraint that kills a naive local-sandbox build), a Serena-seeded PageRank repo-map (rank the most central symbols by running PageRank over the mounted Serena LSP reference edges, no tree-sitter pipeline and no vector store), Goose-style lead/worker failure-threshold escalation (on N worker failures escalate to the lead for re-decomposition instead of grinding), and a per-edit lint ladder capped at max_reflections=3 (changed-file-only lint, escalate not grind). Triggers include split architect and editor across models, route risky exec to a sandbox, build a repo map without embeddings, cap agent retries, escalate a stuck work-unit, or design the orchestration loop. NOT for wiring the Codex fan-out channel itself (use multi-ai-mingle) or picking among finished candidate patches (use execution-grounded-bon).
bmad-workflow-builder
Builds workflows and skills through conversational discovery and validates existing ones. Use when the user requests to "build a workflow", "modify a workflow", "quality check workflow", or "optimize skill".
multi-ai-mingle
Use when the user wants seamless local collaboration between Claude Code and OpenAI Codex on the same machine - fanning out parallel Codex coding threads from Claude as native MCP tool calls via the official codex mcp-server, and running a live symmetric two-way chat and file-lease channel between the two agents through mcp_agent_mail. Covers mounting codex-as-MCP at user scope, standing up the shared agent-mail http channel, verifying it before use, and the sequenced adoption waves. Triggers include mingle Claude and Codex, run Codex from Claude in parallel, set up the agent mail channel, replace AgentBridge, or coordinate the two local AI agents. Not for single-agent work or remote or cloud multi-agent orchestration.
bmad-builder-setup
Sets up BMad Builder module in a project. Use when the user requests to 'install bmb module', 'configure bmad builder', or 'setup bmad builder'.
media-dub
Use when producing a local DUB of a video — re-voicing the speech into another language (or restyling it in the same language) while KEEPING the original speaker's voice, fully offline. The safe-ship engine is Chatterbox-Multilingual (Resemble AI, MIT, 23-language zero-shot voice clone); the pipeline rides the existing media_studio substrate (Parakeet/whisper ASR -> CTC word-alignment -> ffmpeg `atempo` duration-fit -> `sidechaincompress` ducking -> mux). Covers the `dub.start` RPC, the ASR->translate->synth->fit->assemble->duck+mux stages, the per-segment `atempo` time-stretch that lands each clip in its source slot, the full-length VO track assembled with `adelay`+`amix`, and the LICENSE gate: an optional IndexTTS2 native-duration-control engine and an optional LatentSync lip-sync are COMMERCIAL-embed-gated on the `dub-license` needsDecision — the gate runs in SHADOW mode (observe + log, never blocks), and a commercial build defaults to Chatterbox-only unless the owner explicitly opts in (`allowGated`). Do NOT use for: pure TTS narration with no source voice to clone (use the `speech` skill); muxing an ALREADY-finished dub track (use audio-extract-mix); transcription/subtitle generation only (use transcribe/caption); or lip-sync without a dub. Source of truth: media_studio/features/dub.py.
bmad-agent-builder
Builds, edit or validate Agent Skill through conversational discovery. Use when the user requests to "Create an Agent", "Optimize an Agent" or "Edit an Agent".
wds-agent-saga-analyst
Strategic business analyst and product discovery partner for WDS. Use when the user asks to talk to Saga or requests the WDS analyst.
mcp-fleet-curation
Use when plugins, skills, or MCP servers keep REAPPEARING after you disable them, when de-bloating the Claude Code catalogue, or when the per-turn tool/token count creeps back up. The permanent cure - disable at USER scope (claude plugin disable name@mkt --scope user), NOT the UI toggle (which writes project scope and returns when the cwd changes) - plus periodic mcp_doctor audit to catch drift. Covers the two root causes (install = enabled-by-default, and scope mismatch), the reversible park/restore flow for MCP servers in the .mcp.json files, and the catalogue-tax model behind the bloat.
wds-agent-freya-ux
Strategic UX designer and design thinking partner for WDS. Use when the user asks to talk to Freya or requests the WDS designer.
llm-decompile-verify
Use when recovering readable, COMPILABLE C from a stripped binary or a decompiler's pseudo-C and you refuse to trust the model's own claim that the decompilation is correct. The decompilation is only "done" when the regenerated source RECOMPILES and RE-EXECUTES identically to the original on the same inputs (the recompile-verify round-trip). The model runs on HF Inference or cloud-hub GPU; the compile-and-run oracle runs in a cloud-hub Modal sandbox or locally. NOT for the table-stakes "LLM reads Ghidra/IDA pseudo-C" flow (Claude+ghidra-mcp already do that) and NOT for pure static mapping (use game-re-binary) - the gem here is the anti-faking round-trip gate, not the reading. Cite SK2Decompile (arXiv 2509.22114) and LLM4Decompile.
wds-8-product-evolution
Brownfield improvements — the full WDS pipeline in miniature for existing products
imagegen
Use when the user asks to generate or edit images via the OpenAI Image API (for example: generate image, edit/inpaint/mask, background removal or replacement, transparent background, product shots, concept art, covers, or batch variants); run the bundled CLI (`scripts/image_gen.py`) and require `OPENAI_API_KEY` for live calls.
wds-7-design-system
Create, import, browse, and maintain design system components and tokens