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
legal-review
Structures a systematic legal and compliance review of contracts, policies, or technical decisions — identifying key clauses, risks, and questions for counsel. Use when asked to review a contract, check compliance implications, flag legal risks in a technical design, or prepare for a legal consultation. Always surfaces work for qualified counsel; never replaces legal advice.
estimation-planning
Software estimation and planning — story points, velocity, scope negotiation, technical debt budgeting, and statistical forecasting. Use when estimating effort, planning a sprint, negotiating scope with stakeholders, budgeting for technical debt, or forecasting completion dates.
google-workspace
Gmail, Calendar, Drive, Contacts, Sheets, and Docs integration via gws CLI (googleworkspace/cli). Uses OAuth2 with automatic token refresh via bridge script. Requires gws binary.
email-drafter
Drafts clear, professional emails ready to copy and paste — from a brief description of the situation, audience, and intent. Use when asked to write, draft, or compose an email, reply to a message, or suggest how to phrase something professionally.
documentation
Writing effective project documentation — READMEs, API doc comments, and inline comments that explain why. Use when writing or improving documentation — READMEs, doc comments, inline comments, or when code lacks explanation for non-obvious behavior.
devils-advocate
Systematically challenges a plan, decision, or proposal — identifying assumptions, failure modes, and stronger alternatives. Use when asked to critique a plan, stress-test an idea, find weaknesses in a proposal, or when the user says "what am I missing" or "argue against this."
board-of-advisors
Simulates a panel of expert advisors who each evaluate a problem from their distinct perspective — surfacing blind spots and generating more diverse thinking than a single viewpoint. Use when asked to get multiple perspectives, consult experts, think through a decision from different angles, or when a problem benefits from specialist input.
workitem-management
Create, transition, query, and link WorkItems — tasks, stories, bugs, epics. Use when the user asks to add, update, query, or link work items — backlog items, tasks, bugs, user stories, or epics.
task-router
Route a task description to the right processkit skill, project-specific process override, and MCP tool in a single deterministic call — use this at the start of every processkit domain task.
status-update-writer
Generates clear, structured status updates for stakeholders — from project context, backlog state, or a brief description of recent work. Use when asked to write a status update, weekly report, progress summary, or team briefing about a project or initiative.
state-machine-management
Define and customize StateMachine entities — the state/transition graphs used by WorkItems, DecisionRecords, and other primitives with lifecycle. Use when a project needs to customize the default state machine for a primitive (add states, change transitions) or define a state machine for a custom entity type.
unsloth
Expert guidance for fast fine-tuning with Unsloth - 2-5x faster training, 50-80% less memory, LoRA/QLoRA optimization
standup-context
Writes a structured standup update as a session.standup LogEntry — capturing what was done, what is in progress, what comes next, and any blockers. Use when the user says "write a standup", "standup update", "what did I do today", "write a status update for the team", or at the end of a session on projects that run daily standups.
fine-tuning-with-trl
Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and reward model training. Use when need RLHF, align model with preferences, or train from human feedback. Works with HuggingFace Transformers.
skill-reviewer
Audit an existing processkit skill against the 5 Skill Killers, the Agent Skills standard, and Anthropic's troubleshooting playbook, producing a categorized findings report and draft Gotchas.
pytorch-fsdp
Expert guidance for Fully Sharded Data Parallel training with PyTorch FSDP - parameter sharding, mixed precision, CPU offloading, FSDP2
skill-gate
Session-start meta-skill that enforces the 1% rule: before any action on a processkit entity or context/ file, if there is even a 1% chance a processkit skill applies, you MUST check skill-finder first. Load this skill at the start of every session that may involve processkit work. Use when beginning a new task, when unsure which skill applies, or when you are about to edit context/, create an entity, or run an MCP tool.
peft-fine-tuning
Parameter-efficient fine-tuning for LLMs using LoRA, QLoRA, and 25+ methods. Use when fine-tuning large models (7B-70B) with limited GPU memory, when you need to train <1% of parameters with minimal accuracy loss, or for multi-adapter serving. HuggingFace's official library integrated with transformers ecosystem.