Skills
97,725 packaged capabilities and playbooks for agents. Install any of them with one command.
Agent workflows
Auditable agent team: a BMAD fully automated R&D pipeline orchestrator. Orchestrates 9 specialized Agents (PM, Architect, UI Designer, Tech Lead, Scrum Master, Frontend, Backend, QA, DevOps) from requirements to deployment, with event sourcing + non-bypassable gates + deterministic evals at every step — verifiable that tests really run and gates really pass. Supports single-stage slice commands (/boss:plan /review /qa /ship) and a pure Markdown fallback without the CLI. Triggers (full pipeline): 'boss mode', '/boss', '全自动开发', '从需求到部署', '帮我做一个', 'build this', 'ship it', '全流程', '自动化开发', '一键开发', 'start a project', 'new feature' Triggers (single-stage slices, most commonly used on existing projects): - /boss:review — '帮我评审', '代码评审', '技术评审', '看看这段代码有什么问题', '风险评估', 'review this', 'review my PR', 'code review' - /boss:qa — '跑一下测试', '补测试', '检查质量', '测试覆盖够不够', '门禁过了吗', 'run the tests', 'check quality', 'add tests' - /boss:plan — '先出个方案', '写个 PRD', '设计一下架构', 'write a spec', 'plan this feature' - /boss:ship — '可以发布了吗', '部署检查', 'ready to ship', 'deploy checks' Does NOT trigger: - Single-file changes or simple bug fixes (just edit directly) - Pure code reading or explanation (use the read tool) - Repeated invocation while a pipeline is already running - Extremely small tasks (estimated <30 minutes of manual work, no PRD/architecture/gate records needed) Output: complete project code + PRD/architecture/UI/test/deployment documents, written to the .boss/<feature>/ directory
Brev instance operating guidance for NeMo-RL agents working in /home/ubuntu/RL with limited workspace disk, a larger /ephemeral volume, and optional /home/ubuntu/RL/.env secrets. Use when running auto-research campaigns, experiments, training jobs, model or dataset downloads, shared cache-heavy commands, log-producing runs, checkpoint generation, W&B or Hugging Face authenticated workflows, or any workflow that may create large files on Brev.
Use when the agent wants to define, list, inspect, or execute GUI macros via the OpenClaw Macro System CLI. Macros are parameterized, CLI-callable workflows — the agent invokes `macro run <name>` and the system handles backend routing (plugin, file transform, accessibility, compiled GUI replay).
Orchestrate Claude Opus and Codex GPT-5.5 worker TUIs from a controller agent in the current cmux workspace. Use when delegating bounded tasks to worker panes, running plan-first delegations (Claude Code plan mode, Codex Plan mode), monitoring and tracking worker state, or verifying worker reports. Not for end-user cmux control (see cmux, cmux-workspace) — this drives agent workers.
Create a new skill in the current repository. Use when the user wants to create/add a new skill, or mentions creating a skill from scratch. This skill follows the workflow defined in .agents/skills/README.md and helps scaffold, validate, and sync new skills.
System prompt design and instruction crafting for AI agents. Covers persona definition, constraint specification, output formatting, tool use instructions, multi-turn behavior, and guardrail design. Use when writing system prompts for AI products, designing agent personas, specifying behavioral constraints, crafting tool-use instructions, or building multi-step agent workflows.
Run and interpret one safe, bounded Kitaru replay comparison against an accepted cohort and exact evaluators. Use when a user wants to replay a cohort, test or compare a model, prompt, system prompt, parameter, agent version, or tool policy, supervise an experiment run, determine whether one candidate helped, or ask for one bounded change worth testing.
Use this skill for multi-step, spec-driven, or acceptance-gated implementation work in a GitHub repository where specs live as GitHub Issues. Routes work through Plan, Build, and Verify phases, favors demonstrable user-facing vertical slices over waterfall epics, manages the roadmap through issue labels and sub-issues, and triages and grooms the backlog. Also handles migration from file-based specs to GitHub Issues.
Break a bloated, monolithic agent system prompt into a lean stable core plus extracted Ratel-style skills. Use when a prompt is too long or does too many things, to extract or factor procedures into skills, decompose or split a prompt, cut per-turn token cost, or `/ratel-decompose-prompt`. Writes a markdown decomposition plan (lean core, skill inventory, per-skill SKILL definitions, registration wiring) to <repo>/.ratel/ without rewriting the prompt or code. Reached from /ratel-assessment's Prompt Decomposition dimension; pairs with /ratel-tune-definitions.
Turn a finished, failed, or disappointing work cycle into portable lessons, anti-patterns, quality gates, and next-cycle vocabulary. Use after a build, QA pass, demo, user complaint, or abandoned attempt when the useful output is what the next pass must learn rather than the code itself.
Help users discover and select the right Salesforce skill for their task. Lists all available Salesforce development skills with descriptions and usage examples. Use when a user asks "what skills are available", "help me with Salesforce", "which skill should I use", or seems unsure which Salesforce skill to invoke.
Keep a reviewable decision trail for long-running or unattended work: a TSV log with one row per decision (what, why, evidence, result). Local by default; commit it when a reviewer needs the trail to trust the result. Use for /show-me-your-work, autonomous or multi-phase runs, or work a human reviews after stepping away.
Use when the user wants to build a CLI tool to automate browser interactions on a specific website using kimi-webbridge. Invoke when user says "create a CLI for X site", "build a tool to automate X", or wants to control a website programmatically via an AI agent.
Use for tasks that require temporary files, including during another workflow. Also use before creating, inspecting, moving, removing, or reusing agent task directories such as `.agent-*`, and when they appear during repository discovery or validation. Don’t use for tasks limited to durable project outputs, programs’ internal temporary resources, or tool-managed caches.
Design patterns for agentic AI: multi-agent orchestration UX, generative UI, RAG interfaces, LLM hallucination guardrails, confidence indicators, and trust calibration. Use when designing AI copilots or autonomous agents, adding verification affordances, or auditing an AI feature for safety, control, and consent.
Install Alibaba Cloud Agent Toolkit end-to-end: verify and set up prerequisites (uv, Alibaba Cloud CLI, authentication, CLI plugins, MCP Server Core, bearer token exchange), then install the toolkit via the current client's supported plugin mechanism. Use when: alibabacloud agent toolkit install, environment check, prerequisite check, setup alibabacloud, plugin install, toolkit setup, mcp core setup, aliyun CLI setup.
Turn a vague idea or task into a confirmed Working Brief by interviewing the user in rounds — every question that can be asked now, numbered, each with a recommended answer — then hand the brief to whoever does the work. Use when the user invokes Ask Me, wants to clarify scope or requirements before work begins (เคลียร์โจทย์ วางขอบเขต ทำ brief), or asks to execute a WORKING-BRIEF.md.
Use to develop a feature or bug fix end-to-end through a TDD git-worktree workflow — interactively (developer-led) or unattended (autonomous-mode, driven by the dispatcher). Triggers on phrases like "implement issue #N", "fix this bug", "add a feature", "create a worktree", "write test cases", "push and open a PR", "check CI", "address review comments", "resolve review threads", "/q review", "/codex review", "implement this autonomously", or any partial step in the design → worktree → tests → implement → verify → review → PR → CI → E2E lifecycle. Interactive mode asks for decisions; autonomous mode makes decisions per autonomous-mode.md and posts progress comments to the GitHub issue.
Use to perform an end-to-end PR review and reach an approve/request-changes verdict — including verifying acceptance criteria, running E2E tests via browser automation, resolving merge conflicts, and (when verdict passes) merging the PR. Triggers on phrases like "review this PR", "decide whether to approve and merge", "run E2E verification", "resolve merge conflicts on PR #N", or when the dispatcher hands off a PR labeled `pending-review` / `reviewing` for autonomous review. Distinct from in-flight dev-side self-review (that lives in autonomous-dev's pr-review step).
Build a Loop for PostHog Desktop: a workflow that creates an AI task each time its trigger fires, optionally followed by a Slack or email notification with the task's result. Use when asked to create, set up, or change a loop, a recurring agent, a scheduled task, or an automation that runs an AI task when a GitHub, Slack, or PostHog event happens. Covers the exact graph, trigger configs, schedule presets, the notify step, structured output, and the test-run steps.
Internal downstream skill for ctf-sandbox-orchestrator. CTF-sandbox workflow for AI-agent, prompt-injection, MCP or toolchain, cloud, container, CI/CD, and supply-chain challenges. Use when the user asks to analyze prompt-to-tool flows, retrieval poisoning, mounted secrets, deployment drift, runtime-vs-manifest mismatches, registry provenance, or CI-produced artifacts under sandbox assumptions. Use only after `$ctf-sandbox-orchestrator` has already established sandbox assumptions and routed here.
Internal downstream skill for ctf-sandbox-orchestrator. CTF-sandbox workflow for prompt-injection, retrieval poisoning, memory contamination, planner drift, MCP or tool-boundary abuse, and agent exfiltration challenges. Use when the user asks to analyze prompt injection, retrieval poisoning, memory contamination, planner drift, tool-argument corruption, or secret exposure caused by an agent chain. Use only after `$ctf-sandbox-orchestrator` has already established sandbox assumptions and routed here.
Shorten a prompt, specification, plan, skill, note, or brain dump without changing its meaning or behaviour. Use when asked to compress, shorten, simplify, tighten, remove repetition, or use fewer tokens. Preserve facts, constraints, examples, edge cases, and checks that affect the result.
CodeStable usage feedback loop. Use only when the user explicitly asks to record or report that the cs skill went off track, a tool failed, the rules were unclear, or the agent was corrected; collect local evidence and prepare a confirmable public issue.
Run repeated build -> QA -> retro -> re0-work cycles while preserving learning and letting code die. Use for long agentic projects where progress must be measured by quality-cleared templates, reusable modules, and eliminated anti-patterns rather than hours spent or features accumulated.
Select a supported Kitaru adapter or build a project-local one for an unsupported Python or TypeScript agent framework. Use when a user wants to record or replay framework-native agent runs in Kitaru, needs a custom adapter, has no supported Kitaru integration for their framework, or needs to assess whether public framework hooks and the installed Kitaru SDK can support an adapter.
Build and validate a custom Kitaru trace importer when a provider, observability platform, export format, or agent framework has no suitable built-in importer. Use when a user wants to map provider traces into Kitaru sessions and nodes, add bounded provider API fetching, join per-turn traces into longer sessions, preserve incomplete or failed trace evidence, choose script or package installation, test importer fidelity, register an importer version, import a bounded sample, or diagnose a partial import.
SpecStory Lore - mine your SpecStory coding histories (any agent - Claude Code, Codex, Cursor, Gemini, and more) into a persistent corpus, surface your reproducible workflows with corroborated evidence, and interactively forge the chosen ones into skills installed across all your agent harnesses. Use when the user wants to turn past AI coding sessions into reusable skills, asks "what could I make into a skill", "mine my lore", "forge skills from my history", or points at a .specstory/history directory.
Turn an external tech/AI source (YouTube video, podcast, blog article, tweet or thread) into a de-duplicated improvement roadmap for the MEL system and Claude Code setup: fetch the content, extract concrete techniques, inventory what the system already does, classify each idea reuse/improve/rebuild on an Adopt/Trial/Assess/Hold ring, and write the roadmap file with a mandatory What-NOT-to-build section. Use for "analyse this video/article/talk/thread for my system", "turn this into a system-improvement plan", or a pasted URL with improvement intent. Not for MEL research video files needing the transcription+consent pipeline (video-content-analysis), mining internal QA logs (/improve-system), inspecting the system itself (/system-audit, /grade-system), academic evidence review (evidence-synthesis), or writing the LinkedIn piece (linkedin-field-note).
Add Opik tracing to an existing app and verify a real trace lands. Installs the Opik package, detects the language and LLM framework, adds the minimum tracing, runs a safe representative path, confirms a trace in Opik, and returns the trace link. Use for “instrument my code”, “add opik tracing”, “add observability”, “trace my agent”. Not for building a new app from scratch, or a review-only pass with no code changes.
Search context data (memories, skills and resources) from OpenViking Context Database (aka. ov). Trigger this tool when 1. you need information that might be stored as memories, skills or resources on OpenViking; 2. searching files or knowledge is explicitly requested; 3. you see a `search context`, `search openviking`, `search ov` request.
Planning methodology and session management for software projects. Covers markdown plan creation, iterative refinement, multi-model blending, phase-based implementation docs, and cross-session progress tracking. Use when starting new projects, adding major features, breaking large work into phases, resuming after context clears, or managing multi-phase implementations.
Prompt patterns for consistent UI generation: stating design intent precisely, component specification formats, few-shot examples, and structured refinement loops. Use when UI output from an LLM is inconsistent or generic, or when building reusable prompt templates for a design system.
This skill should be used when users request help optimizing, improving, or refining their prompts or instructions for AI models. Use this skill when users provide vague, unclear, or poorly structured prompts and need assistance transforming them into clear, effective, and well-structured instructions that AI models can better understand and execute. This skill applies comprehensive prompt engineering best practices to enhance prompt quality, clarity, and effectiveness.
Audit a codebase's tool and skill definitions against a failure-mode rubric and write a markdown tuning plan with per-definition before/after rewrites that make them retrievable by Ratel's hybrid lexical + semantic index and usable by the model. Use when the model picks the wrong tool, tools never get picked, descriptions are too long/vague/anemic, to fix near-duplicates, tighten schemas/enums/param names, clean up the tool catalog, or `/ratel-tune-definitions`. Fetches live Ratel docs; writes to <repo>/.ratel/ without editing tool or skill code. The fix /ratel-assessment's Definition Quality dimension routes to; runs after /ratel-decompose-prompt.
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.