Agent Workflow Claude Skills
4 121 agent workflow skills from 665 authors. They configure the agents themselves: memory, prompts, context and other skills. Half of them fit into 1 845 tokens or less — that is what one costs your context window when the agent loads it. 778 ship runnable scripts rather than instructions alone. 5 of them cannot work without an MCP server, most often task. We also found 541 copies of these same skills sitting in other people's repositories — counted once here, not 541 times.
4 121 unique 665 authors 2 767 updated this month 506 from vendors
>- Agent-native DeepRefine refinement loop — same control flow as DeepRefine.refine(), graphify search instead of FAISS, session LLM, dry-run review before approved graph writes.
>- Claude Code adapter for the DeepRefine agent-native refinement loop. Use when the user invokes /deeprefine, or asks to refine, diagnose, review, or apply changes to a Graphify / LLM-Wiki knowledge graph. Must follow the canonical DeepRefine skill rules and stop for review before graph writes.
Generate clinical trial protocols for medical devices or drugs. This skill should be used when users say \"Create a clinical trial protocol\", \"Generate protocol for [device/drug]\", \"Help me design a clinical study\", \"Research similar trials for [intervention]\", or when developing FDA submission documentation for investigational products.
Automate payer review of prior authorization (PA) requests. This skill should be used when users say \"Review this PA request\", \"Process prior authorization for [procedure]\", \"Assess medical necessity\", \"Generate PA decision\", or when processing clinical documentation for coverage policy validation and authorization decisions.
Master advanced prompt engineering techniques to maximize LLM performance, reliability, and controllability in production. Use when optimizing prompts, improving LLM outputs, or designing production prompt templates.
This skill provides mandatory safety guidelines for coding agents. It should be used at the start of any coding session and whenever performing file operations, executing code, handling credentials, or interacting with external systems. The protocol defines what actions are permitted, restricted, or prohibited to ensure secure, reliable, and ethical agent behavior.
Guide for creating effective skills. This skill should be used when users want to create a new skill (or update an existing skill) that extends Gemini CLI's capabilities with specialized knowledge, workflows, or tool integrations.
Replace with description of the skill and when Claude should use it.
Audit whether an AI agent can autonomously close the loop on problems in a given area — from discovering a symptom to verifying a fix — without human intervention. Evaluates documentation, diagnostic tools, commands, logs, and flows for completeness and actionability. Generates a gap-focused report with ratings. Use this skill whenever someone wants to assess debugging readiness, check if docs are agent-sufficient, audit a workflow for autonomous solvability, evaluate operational tooling coverage, or wants to know 'could an agent fix this on its own?' Triggers on: 'loop audit', 'audit this flow', 'is this debuggable', 'agent readiness', 'can an agent solve this', 'autonomous debugging check', or /hk-autonomy-audit.
Consults existing breadcrumb analysis docs before exploring the codebase. Use this skill whenever the user asks about how a flow works, where something happens in the code, how to debug or test a specific area, what files are involved in a feature, or needs to understand the path data takes through the system. Also trigger when the user mentions 'where does X happen', 'how does Y work', 'trace this', 'what files handle Z', 'how to test this flow', 'debug this area', or when you're about to spawn multiple exploration subagents to understand a cross-cutting flow. Even if the user doesn't explicitly ask — if the task requires understanding how multiple layers connect (frontend → backend → worker → CLI), check breadcrumbs first. This is cheaper and more accurate than re-discovering the same information through code search.
Run comprehensive agent-native architecture review with scored principles. Audits a codebase against 8 agent-native architecture principles (Action Parity, Tools as Primitives, Context Injection, Shared Workspace, CRUD Completeness, UI Integration, Capability Discovery, Prompt-Native Features) by launching parallel sub-agents and producing a scored report. Use when the user wants to evaluate how agent-friendly their architecture is, or audit specific principles. Triggers on: 'agent native audit', 'architecture review', 'how agent-friendly is this', or /hk-arch-audit.
Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy. Triggers on: 'create a skill', 'new skill', 'make a skill for', 'improve this skill', 'test this skill', 'skill eval', 'optimize skill description', or /hk-skill-creator.
Benchmark agent skills by generating eval cases, comparing skill-guided vs baseline runs, and recording artifacts under skills-test/{skill-name}.
Keep repository-level Claude and Codex agent instructions compatible by creating safe symlinks between CLAUDE.md and AGENTS.md, and between .claude/skills/* and .agents/skills/*. Use when a repository should work with both Claude and Codex, when one agent instruction file exists but the other is missing, when migrating agent skills between .claude and .agents directories, or when asked to improve Claude/Codex compatibility for a repo.
Route GPT-5.6 tiers, reasoning effort, and subagents to minimize accepted-result cost subject to an explicit quality floor.
Route Claude 5 tiers (Haiku/Sonnet 5/Opus 5/Fable 5), effort, and subagents for cost-effective accepted results, and handle Claude 5 prompting, long-running execution, API behavior, refusals, and fallback.
Audit and clean the persistent context that feeds a recurring autonomous agent loop — memory files and indexes, scheduled-task / automation prompts, and CLAUDE.md / AGENTS.md — so the loop stops degrading into a self-reinforcing echo chamber. Works for Claude Code /loop crons and Codex automations.
This skill should be used when the user wants to install or configure Pi Agent (@earendil-works/pi-coding-agent) with DeepSeek (built-in), Ant-Ling Ring-2.6-1T (single-model custom provider), and ZenMux (multi-model OpenAI-compatible aggregator including Gemini 3.5 Flash, inclusionAI Ling-3.0-flash, Claude, GPT), including auth, models.json, settings.json, a curated extension set, and known-pitfall fixes. Triggers on "配置 pi"、"setup pi agent"、"pi 装一下"、"配 ring/deepseek/gemini/zenmux/ling 到 pi".
Find Agent Skills on skills.sh and adopt only the ones that pass a security audit. Use when the user wants to discover, search for, evaluate, or install a third-party skill ("find a skill for X", "is this skill safe to install", "search skills.sh", "what skill should I use for Y"). Always runs the skills.sh security audit before recommending or installing anything, and refuses to adopt skills that fail the audit.
将具体执行委派给 pi coding agent + ZenMux 便宜执行模型（默认强档 deepseek/deepseek-v4-pro：1M 上下文、推理型；廉价快档 inclusionai/ling-3.0-flash：256K、非思考、更省），Claude 负责任务-模型匹配判断、任务书编写、驱动（一次性 pi -p 或 tmux 交互长程）、硬超时重试、独立验收与失败裁决。含双模型选型、3D 素材来源、Spec-Driven、强类型可验证节点、推理档位开关等最佳实践。触发词：pi 委派、把执行外包给便宜模型、deepseek、deepseek-v4-pro、v4 pro 执行、ling、ling-3.0-flash、ling flash 执行、pi sub agent
Compact the current conversation into a handoff document so a fresh agent can continue the work.
Plan a huge chunk of work — more than one agent session can hold — as a shared map of investigation tickets, and resolve them one at a time until the way to the destination is clear.
> Silently restructures the user's natural-language prompt into the format the model CURRENTLY running this skill handles best, then answers. On activation it identifies which model family is executing it (Claude, GPT, Gemini, Llama, DeepSeek, Mistral, Qwen, Grok, Perplexity, Kimi, GLM, Command, Nova, or Phi) and loads that one model's official strategy — so the optimization always matches the model that actually runs it. Activate with /prompt-refine. Use when users want better answers without learning prompt engineering.
Share a Claude Code session to pastila.nl and return a viewer link. Shares the current session by default, or a session specified as a transcript path, a session id, or a project. Use when the user asks to share, publish, or get a link to this conversation or to a session/transcript.
Context preservation with cross-session memory — hooks prompt compaction at high-impact moments, siblings enrich what gets saved
Universal MCP client for connecting to any MCP server with progressive disclosure. Wraps MCP servers as skills to avoid context window bloat from tool definitions. Use when interacting with external MCP servers (Zapier, Sequential Thinking, GitHub, filesystem, etc.), listing available tools, or executing MCP tool calls. Triggers on requests like "connect to Zapier", "use MCP server", "list MCP tools", "call Zapier action", "use sequential thinking", or any MCP server interaction.
Load when a task needs read-only internet research or content retrieval through an already installed Agent Reach CLI; diagnose available backends first, never install, upgrade, authenticate, configure, or copy credentials on the user's behalf.
Load when the user asks for private agent interaction audits from local Codex/Claude Code traces, task profiles, tool-call decisions, collaboration bottlenecks, prompt/rule or automation-log audits, closeout retrospectives, or evidence-backed Skill/project-rule/Eval/SOP/OKF improvements.
Load when a task needs agent run, tool loop, context drift, or recoverable harness/tool failure debugging; use diagnose for product/runtime bugs.
Load when agent pauses to report relatively complex information needing Chinese-first clear complex communication, alignment, multi-option choice, status/incident, long-task fact ledgers, implementation plans, reviews, maps, explainers, evidence, risks, validation, handoff; choose plain text/Markdown/visual Markdown/HTML by decision cost; do not load merely because answer is long; skip trivial chat/bundled apps.
Load when a task needs to hand off current work, compact a conversation for another agent/session, or create restart notes; do not load for visual HTML repo reports.
Load when a task needs MCP server design, build, review, testing, tool schema, resource, or prompt work; skip ordinary REST clients and non-MCP integrations.
Load when creating, updating, adapting, or evaluating a standard agent skill; do not load for ordinary documentation edits or one-off prompt advice.
Create or update a complete repository skill from a user's idea, including the workflow instructions, references, scripts or assets, agent metadata, skill-card artwork, cinematic banner artwork, README links, discovery metadata, and validation. Use when the user asks to create a new skill, add a skill to this collection, turn a workflow into a reusable skill, or make a skill's documentation and artwork consistent with the repository.
Profile and debug Hermes sessions from their JSONL transcripts. Find a session and its subagents, build a queryable event table, summarize the work as a hierarchical table of contents, break down wall time, inference, tools, tokens, and estimated cost per agent, identify errors and improvement opportunities, and export a shareable Perfetto trace. Use when a user wants to inspect what a Hermes session did, debug agent or subagent activity, understand session cost or latency, create a session timeline, generate a trace, or learn how to improve the next agent run.
Plan and build MCP servers with agent-friendly tools, schemas, error handling, and evaluation. Use when creating or refactoring MCP integrations.
Create or update agent skills with concise SKILL.md instructions, bundled resources, agent metadata, validation, and packaging. Use when building or refactoring skills.
Operating-instructions dispatcher for the openclaw-coder autonomous-programmer workspace. Routes every user message by surface — thread → working session, channel/DM → channel handling — and carries the global rules. The workspace AGENTS.md loads this skill first on every user message.
Proactively decompose and coordinate substantial development, research, analysis, planning, document, data, and content work with the smallest useful parallel set of custom subagents. Use when independent execution, parallelism, context isolation, or fresh review can improve speed or quality. Keep unresolved decisions and final acceptance in the main thread. Do not use for casual or simple tasks.
Structured self-debugging workflow for AI agent failures using capture, diagnosis, contained recovery, and introspection reports.
Build an evidence-backed ecc install plan for a specific repo by sorting skills, commands, rules, hooks, and extras into DAILY vs LIBRARY buckets using parallel repo-aware review passes. Use when ecc should be trimmed to what a project actually needs instead of loading the full bundle.
Multi-agent orchestration using dmux (tmux pane manager for AI agents). Patterns for parallel agent workflows across OpenAI Codex, Codex, OpenCode, and other harnesses. Use when running multiple agent sessions in parallel or coordinating multi-agent development workflows.
Build MCP servers with Node/TypeScript SDK — tools, resources, prompts, Zod validation, stdio vs Streamable HTTP. Use Context7 or official MCP docs for latest API.
Build MCP servers with Node/TypeScript SDK — tools, resources, prompts, Zod validation, stdio vs Streamable HTTP. Use Context7 or official MCP docs for latest API.
Structured self-debugging workflow for AI agent failures using capture, diagnosis, contained recovery, and introspection reports.
Build an evidence-backed ecc install plan for a specific repo by sorting skills, commands, rules, hooks, and extras into DAILY vs LIBRARY buckets using parallel repo-aware review passes. Use when ecc should be trimmed to what a project actually needs instead of loading the full bundle.
Multi-agent orchestration using dmux (tmux pane manager for AI agents). Patterns for parallel agent workflows across OpenAI Codex, Codex, OpenCode, and other harnesses. Use when running multiple agent sessions in parallel or coordinating multi-agent development workflows.
日本語翻訳：このファイルは mcp-server-patterns 用の日本語翻訳が必要です