Agent Skills
Skill Library
46 copyable SKILL.md files. Each follows the Agent Skills standard: a clear trigger, exact steps, and the pitfalls that bite. Drop any one into a Claude Code project.
46
Skills
8
Categories
SKILL.md
Standard
Skills of skills
Full guideHow to Write a Great Skill
Use when writing a new SKILL.md from scratch or improving an existing one. Covers the four sections every skill needs, how to write a trigger an agent will actually match, what the pitfalls section is really for, and the signs a skill is complete rather than still a draft. Also use when a skill exists but never gets loaded, reads as documentation the model already knows, or is too broad to be useful. If you are authoring a skill, use this skill.
MCP02Skill Testing Checklist
Use after writing a new skill or editing an existing one, to verify it will be loaded at the right time, followed correctly, and not cause problems in adjacent tasks. Covers the trigger probe, the procedure walk-through, the pitfall audit, the scope test, and the staleness check. Also use when a skill loads for the wrong tasks, or is being followed but producing the wrong result. If a skill is about to ship, use this skill.
MCP03Skill Composition Patterns
Use when a task is too large or too multi-domain for a single skill, when you notice overlap between existing skills, or when you want a skill system that routes between skills cleanly. Covers the entry-router, reference-file, and skill-chain patterns, split versus merge decisions, and safe cross-linking without duplicating the repo's agent instructions. Also use when two skills give conflicting advice, or one skill has grown into several jobs. If skills need to work together, use this skill.
MCPWrite the trigger
The description is the whole match surface. Write it as a condition, not a topic.
Exact steps, real pitfalls
Show the actual command. State the actual failure. "Be careful" is not a pitfall.
Test before shipping
Probe the trigger, walk the procedure, audit each pitfall. Then watch a live agent use it.
Featured
46 in libraryAgent Fleet Coordination
Use when running more than a couple of subagents against one repo and the hard part has shifted from splitting the work to merging it back safely: who owns which file, how results get reconciled, and how you prove the merged commit actually builds. Covers the orchestrator, finder, and verifier roles, single-writer file ownership, and a collect-then-apply harvest loop. Also use when two agents have overwritten each other, or a merged commit passed review and failed the build. If several agents share a repo, use this skill.
MCP Server Builder
Use when building or extending a Model Context Protocol server that exposes tools to Claude or other agents, especially as a Next.js App Router route with mcp-handler. Covers tool design as prompt engineering, the concrete route wiring, strict zod schemas, Bearer auth, and evaluating tools so the model actually calls them. Also use when a tool exists but the model ignores it, or a tool's schema lets the model pass something unusable. If an agent will call your API, use this skill.
CLAUDE.md Authoring
Use when creating or improving a CLAUDE.md or AGENTS.md so coding agents follow project conventions. Covers which rules are worth stating (the non-obvious ones), which commands matter, and what to leave out so the file does not become a repo dump. Also use when agents keep repeating a mistake the file was supposed to prevent, or the file has gone stale against the code. If agent behavior needs to change, use this skill.
Credit Metering for SaaS
Use when charging for AI or compute-heavy actions in a SaaS with a credit balance rather than per-call Stripe charges. Covers the per-action cost table, an atomic debit, free actions, an owner bypass, and turning an insufficient balance into a pricing redirect rather than a broken action. Also use when a balance can go negative under concurrency, or a failed action still charged the member. If an action costs the member something, use this skill.
Verifying the Committed Tree Before Push
Use before pushing to a deploy branch on a repo where fast iteration or agents leave files uncommitted, so a commit that imports an unstaged module cannot pass local checks and then fail the remote gate. Covers typechecking an isolated worktree of HEAD rather than the dirty working tree. Also use when a build is green locally and fails on the deploy with a module-not-found. If a push can break main, use this skill.
All skills
CommunityNext.js & Web3
Next.js App Router Conventions
Use when creating or restructuring anything under a Next.js app/ directory: adding a route, page, or layout, deciding whether a component is a Server or Client Component, adding "use client", writing a route handler, setting metadata, or moving data fetching. Also use when a build complains that a hook or event handler cannot be used in a Server Component, or when data is fetched in the wrong place. If a file lands under app/, use this skill.
Playwright Visual Verification
Use when you need to confirm a frontend change actually renders in the running app, not just that tests pass and the code compiles. Also use when a change looks right in code but a panel comes up empty, a menu does not open, or the layout shifts on load. Covers driving the page, waiting on real signals rather than sleeps, and asserting on visible state. If a UI change needs proof, use this skill.
E2E Route Testing with Playwright
Use when adding end-to-end coverage to a Next.js app with Playwright, especially a route smoke suite proving every public page renders. Covers config, a data-driven route check, auth-gated routes, and keeping the suite deterministic with real waits instead of sleeps. Also use when a suite is flaky, or a page 500s in production that no test covered. If a route can break without a test noticing, use this skill.
Auth & Data7
Clerk Auth Setup (Next.js)
Use when adding Clerk authentication to a Next.js App Router app, protecting routes behind sign-in, or reading the signed-in user in server code. Also use when auth() throws in a route handler, a protected route lets an unauthenticated request through, or the provider and middleware disagree about which routes are public. If a request needs to know who the user is, use this skill.
Neon + Drizzle Migrations
Use when defining or changing a database schema with Drizzle ORM on Neon Postgres, generating a migration, or applying one to a live database. Also use when the schema file and the deployed database have drifted, a migration fails to apply, or local and production disagree about a column. If a table definition changes, use this skill.
Neon Serverless Client Patterns
Use when wiring the Neon serverless driver into a Next.js app with Drizzle, especially when the build must succeed in environments with no DATABASE_URL. Covers lazy client construction, the HTTP versus pooled tradeoff, and edge versus node runtime. Also use when a build fails at import time on a missing connection string, or a migration needs the direct rather than the pooled URL. If a database client is constructed at module scope, use this skill.
Clerk Dev vs Prod Gotchas
Use when a Next.js app uses domain-locked production Clerk keys, so local development cannot run clerkMiddleware and auth() throws in route handlers. Also use when a dashboard route 500s locally but works deployed, or an API returns 404 where it should return 401. Covers the middleware bypass and a null-safe server helper. If auth works in production but not locally, use this skill.
API Key Auth Design (Hash-Only)
Use when adding programmatic API key auth to a service. Covers show-once issuance, storing only a sha256 hash plus a display prefix instead of the secret, verifying a bearer token by hashing it, and revocation. Also use when a key is currently stored in a form that could be read back, or the UI needs to show a key it must not keep. If a caller authenticates without a session, use this skill.
Deploy & Ops5
Vercel Deploy Debugging
Use when a deploy fails but the local build passes, or a deploy succeeds and the deployed site is wrong. Covers a build that is green locally and red in CI, a type error only the deploy sees, a missing or wrongly scoped environment variable, a file that was never committed, a lockfile or package-manager version mismatch, and a stale build cache. If a deploy log needs reading, use this skill.
Oxlint as a Fast CI Gate
Use when adding oxlint to a JS/TS project as a fast pre-commit and CI gate. Covers config, an autofix pass, the errors-must-be-zero and warnings-advisory split, and how it coexists with a type gate rather than replacing it. Also use when lint is slow enough that people skip it, or CI is failing on advisory warnings. If lint needs to be fast and honest, use this skill.
GitHub Actions for Agent-Written Code
Use when setting up GitHub Actions CI for a repo where coding agents open PRs, so the gate catches the failure modes agents actually produce: an import of a file that was never committed, lockfile drift, and blocked dependency build scripts. Covers reproducing the deploy's install and type gate exactly, and verifying the committed tree rather than the working tree. If an agent's commit can reach main, use this skill.
Request Tracing with Spans
Use when adding lightweight per-request observability to an API, especially an LLM or media route, so you can see where the time went, which model ran, and what the request cost - without a heavyweight tracing vendor. Also use when a route is intermittently slow and nothing records which step was responsible. Covers timed spans, one durable row per request, and best-effort writes that never throw into the caller. If latency or cost needs attributing, use this skill.
Verifying the Committed Tree Before Push
Use before pushing to a deploy branch on a repo where fast iteration or agents leave files uncommitted, so a commit that imports an unstaged module cannot pass local checks and then fail the remote gate. Covers typechecking an isolated worktree of HEAD rather than the dirty working tree. Also use when a build is green locally and fails on the deploy with a module-not-found. If a push can break main, use this skill.
Agents & MCP14
Agent Fleet Coordination
Use when running more than a couple of subagents against one repo and the hard part has shifted from splitting the work to merging it back safely: who owns which file, how results get reconciled, and how you prove the merged commit actually builds. Covers the orchestrator, finder, and verifier roles, single-writer file ownership, and a collect-then-apply harvest loop. Also use when two agents have overwritten each other, or a merged commit passed review and failed the build. If several agents share a repo, use this skill.
MCP Server Builder
Use when building or extending a Model Context Protocol server that exposes tools to Claude or other agents, especially as a Next.js App Router route with mcp-handler. Covers tool design as prompt engineering, the concrete route wiring, strict zod schemas, Bearer auth, and evaluating tools so the model actually calls them. Also use when a tool exists but the model ignores it, or a tool's schema lets the model pass something unusable. If an agent will call your API, use this skill.
Progressive Disclosure for MCP
Use when an MCP server, or any tool surface an agent reads, exposes a collection whose items are too big to dump at once - skills, files, docs, records, design systems. Covers the three-tier lean index to manifest to item pattern so an agent spends context only on what it opens, how to name the tiers, how to shape each payload, and when to collapse the middle tier. Also use when a tool response is blowing out the context window, or an agent must read everything to find one item. If a listing could be large, use this skill.
How to Write a Great Skill
Use when writing a new SKILL.md from scratch or improving an existing one. Covers the four sections every skill needs, how to write a trigger an agent will actually match, what the pitfalls section is really for, and the signs a skill is complete rather than still a draft. Also use when a skill exists but never gets loaded, reads as documentation the model already knows, or is too broad to be useful. If you are authoring a skill, use this skill.
Skill Testing Checklist
Use after writing a new skill or editing an existing one, to verify it will be loaded at the right time, followed correctly, and not cause problems in adjacent tasks. Covers the trigger probe, the procedure walk-through, the pitfall audit, the scope test, and the staleness check. Also use when a skill loads for the wrong tasks, or is being followed but producing the wrong result. If a skill is about to ship, use this skill.
LLM Engineering6
Single-File App Generation
Use when prompting an LLM to emit a complete, self-contained HTML app - all HTML, CSS, and JS in one file - that you will drop into a sandboxed iframe and run: an artifact, a preview pane, a generated micro-app, a code playground. Covers the output contract (one file, pinned CDN versions, module scripts, no build step), the iframe sandbox and CSP that make running untrusted output safe, and size caps. Also use when a generated file looks plausible and renders as a blank frame. If model output will execute in a browser, use this skill.
Refusal and Fallback Handling
Use when a product feature runs on an LLM and must stay usable when the model refuses, returns malformed or invalid JSON, or times out. Also use when users are seeing a raw provider error, a feature dead-ends on a refusal, or a retry loop is making things worse instead of better. If a model failure can reach the user, use this skill.
Prompt Caching Optimization
Use when an LLM feature sends a large, mostly-stable prompt on every request - a long system prompt, tool definitions, retrieved context - and cost or latency needs to come down. Also use when cache hit rates are lower than expected, or a per-request value placed early in the prompt is invalidating the whole prefix. If the same tokens are being paid for repeatedly, use this skill.
Structured Output Extraction
Use when extracting structured, typed data from unstructured text with an LLM - parsing a document, classifying, pulling named fields - and downstream code needs a guaranteed shape. Also use when the model wraps its JSON in prose, omits a required field, or drifts in shape between calls. If a model's output is parsed rather than displayed, use this skill.
Vercel AI Gateway Integration
Use when routing an app's LLM calls through the Vercel AI Gateway with the AI SDK so one key reaches many providers. Covers plain string model ids, a client-safe model whitelist, and streaming. Also use when a model id works against a direct provider but fails through the gateway, or the gateway key is at risk of reaching the client. If a model call crosses a gateway, use this skill.
SEO & Content2
Content Pipeline
Use when taking a topic or a published video through research to a shipped, source-linked, SEO-structured blog post - the full path from raw idea to a post that ranks and stays fair. Covers research, drafting, SEO structure, cross-linking, and the pre-publish checklist. Also use when a published video has no post yet, or a draft is missing its primary sources, internal links, or FAQ. If something is being published, use this skill.
SEO Decision-Intent Posts
Use when writing decision-intent content: a versus comparison, a pricing breakdown, an alternatives list, or a tool-selection guide that needs to rank on search and stay genuinely fair to every option covered. Also use when a draft makes claims with no primary source, reads as promotional for one option, or has no structured FAQ. If a reader is choosing between tools, use this skill.
Design & UI5
Dashboard Design (Vercel + shadcn)
Use when building or editing a member or dashboard surface that should feel like Vercel and shadcn - soft rounded, light, breathing - and NOT like a hard-edged square marketing site. Covers the neutral palette, radii, spacing, sidebar, tabs and card patterns, and the strict bans. Also use when a dashboard page has drifted toward marketing styling, or a new surface does not match the ones beside it. If a page lives behind sign-in, use this skill.
shadcn Component Adaptation
Use when adding a shadcn/ui component to a project with an opinionated design system, so the copied component matches the system rather than the shadcn defaults. Also use when an installed component brings radii, shadows, or colors that fight the design contract, when accessibility was lost while restyling, or when variants have drifted apart across components. If a component is copied in from a registry, use this skill.
Streamdown Chat Rendering
Use when rendering streaming LLM markdown in a chat UI with Streamdown. Covers incomplete-token safety mid-stream, a custom component map for a strict design system, and code-block handling. Also use when a half-arrived token renders as broken markup, a code fence flickers while streaming, or rendered output ignores the design contract. If markdown is rendered before it is complete, use this skill.
Hyperframes
Use for any request to make, create, edit, animate, or render a video, animation, or motion graphic - a promo, an explainer, a captioned clip, a title card, an overlay. HyperFrames renders video from HTML compositions, and this is the entry skill: read it before any other video skill rather than guessing a workflow, and it routes to the right domain skill. Defer only when asked to drive a browser to capture a session, or when another framework is named outright. If a video needs authoring or rendering, use this skill.
Remotion Best Practices
Use when creating or editing a Remotion video in React: setting up a project, animating with useCurrentFrame and interpolate, timing a sequence, loading assets, or rendering the output. Also use when an animation does not move, the render looks different from the preview, an asset fails to load during render, or a composition's duration and frame rate disagree. If a video is written as React components, use this skill.
Billing4
Credit-Metered AI Endpoints
Use when monetizing an AI endpoint - image, voice, video, or LLM generation, or any paid tool call - with a prepaid credit balance instead of per-call Stripe charges. Covers the deduct-on-success ordering, an atomic single-UPDATE spend that cannot go negative under concurrency, a consistent 401/402/429 error envelope, rate limiting, an append-only ledger, and an owner bypass. Also use when a failed call still charged, or two concurrent calls both passed a balance check. If an endpoint costs money to run, use this skill.
Credit Metering for SaaS
Use when charging for AI or compute-heavy actions in a SaaS with a credit balance rather than per-call Stripe charges. Covers the per-action cost table, an atomic debit, free actions, an owner bypass, and turning an insufficient balance into a pricing redirect rather than a broken action. Also use when a balance can go negative under concurrency, or a failed action still charged the member. If an action costs the member something, use this skill.
Autumn Credit Billing
Use when adding Autumn, a billing layer in front of Stripe, to sell a credit balance shared across apps. Covers using the Clerk userId as the Autumn customer id, hosted checkout, verifying Svix-signed webhooks, and granting credits only after re-verifying the purchase. Also use when a webhook grants credits it should not, or a purchase completes without the balance moving. If money buys credits, use this skill.
Owner Bypass for Founder Accounts
Use when a metered SaaS needs founder, team, or demo accounts to run every paid action for free, without scattering internal-account checks across the codebase. Covers an isOwner flag on the balance row, an env-backed owner list, and honoring the bypass inside the atomic debit so no call site has to remember it. Also use when a free account is still being charged, or a bypass was forgotten on a new route. If some accounts must not be billed, use this skill.
Get Smarter About AI Dev
New tutorials, open-source projects, and deep dives on coding agents - delivered weekly.