vercel-labs/open-agents
8 skills404k installs44.9k starsGitHub
Install
npx skills add https://github.com/vercel-labs/open-agentsSkills in this repo
1Agent Browseragent-browser is a Vercel Open Agents skill that wraps a CLI for programmatic browser control—ideal when solo builders need to navigate sites, interact with DOM elements, fill forms, reload flows, capture evidence, or scrape structured data without maintaining a separate Playwright script repo. The documented workflow is deliberately small: open a URL, take an interactive snapshot to obtain element refs, perform clicks and fills by ref, then re-snapshot after navigation or heavy DOM churn. Commands cover full navigation (including connect 9222 for CDP), aliases for goto and quit, and page analysis via snapshot -i. It complements agentic coding loops where the model verifies UI behavior after a build. Complexity sits at intermediate because you must manage browser lifecycle, ref invalidation, and allowed bash tooling boundaries.404kinstalls2Web Animation DesignWeb Animation Design is an agent skill that helps solo and indie builders design and implement web motion that feels intentional rather than decorative noise. It activates whenever you ask about easing curves, duration, spring physics, keyframes, hover microinteractions, page transitions, or why an animation feels janky. The playbook synthesizes practical rules from Emil Kowalski’s Animations on the Web course so you can choose ease-out versus ease-in-out, stagger lists, and entrance or exit timing with confidence. It also steers performance habits such as transform and opacity, will-change discipline, and reduced-motion accessibility. When you want a critique, the skill enforces a structured markdown review format instead of vague feedback. Use it while polishing dashboards, marketing sites, or app shells where small motion cues signal quality and trust.129installs3Plan Modeplan-mode from Vercel open-agents is a journey-wide planning skill for solo builders who want agents to treat each request as a system change rather than an isolated edit. It activates on explicit plan triggers and on non-trivial work: new capabilities, architectural forks, ambiguous requirements, or several equally valid designs. The embedded doctrine pushes analysis of data flow, state ownership, interfaces, lifecycle, failure modes, and observability before any code lands, favoring designs that can be derived from backend events or persisted truth instead of ad-hoc UI state. Scope expands only when widening the change actually simplifies the whole system. For indie operators juggling Claude Code, Cursor, or Codex, plan-mode is the guardrail against quick patches that mortgage future velocity—especially on SaaS and agent-shaped codebases where coupling spreads fast across files.76installs4Code ReviewCode Review is an agent skill from Vercel Open Agents that turns vague “look at my changes” requests into a repeatable review ritual. Solo builders shipping from feature branches use it when they want an agent to behave like a diligent reviewer: infer whether the target is working tree, a commit, a branch comparison, or a GitHub pull request, then gather the right git or gh output. The skill stresses that patches alone mislead—you must load full files to see patterns, auth paths, and error handling around the edit. That makes it valuable right before merge during Ship, and again during Operate when you audit hotfixes or incident branches. Output is actionable commentary anchored in repository context rather than generic style nitpicks. Pair it with your existing test and security skills; it does not replace CI but complements human or agent judgment at review time.33installs5Ai Sdkai-sdk is a reference skill from Vercel Labs that teaches agents how to call models through Vercel AI Gateway using the AI SDK—the fastest path for solo builders who want one HTTP surface instead of juggling per-vendor SDK keys. It covers authentication patterns (OIDC on Vercel, API keys locally), the two import styles for the gateway provider, and the operational rule to pull the current model catalog via curl before coding. That last constraint matters because model slugs change often and hallucinated IDs break builds quietly. Use it when you scaffold chat features, summarization endpoints, or agent loops in Next.js or Node backends destined for Vercel. It is intentionally narrow: gateway transport and SDK calls, not prompt design or evaluation harnesses.27installs6WorkflowWorkflow is an agent skill from Vercel Labs for solo and indie builders who need orchestration that outlasts a single process lifetime. It steers your coding agent toward Vercel’s Workflow DevKit when you mention durable functions, resumable runs, step-based coordination, or the workflow devkit itself. The skill’s core rule is to treat bundled in-repo documentation as source of truth—search node_modules/workflow/docs for setup guides, foundations like workflows-and-steps and hooks, API references for sleep and hooks, client methods to start and resume runs, AI SDK integration notes, and error codes—rather than relying on stale model knowledge. That matters for SaaS backends, agent pipelines, and API products where a deploy or crash must not lose in-flight work. Use it while designing multi-step flows with external waits, retries, and observability-friendly run IDs. It pairs naturally with Node frameworks on Vercel but the mental model—explicit steps, hooks, fatal errors—applies wherever the kit is installed.25installs7Deploy Open Harnessdeploy-open-harness guides solo builders through forking and running Open Harness on Vercel without guessing which environment variables the app actually reads at runtime. The skill instructs the agent to reconcile README guidance against live paths such as the DB client, JWE encryption helpers, Vercel and GitHub auth routes, Redis, and sandbox configuration so deployment advice matches the repo you have checked out. You choose between a lean deployment and the fuller GitHub-connected agent harness, gather credentials in that order, deploy on Vercel, and validate sign-in and app install flows. Safety rules keep secrets out of chat logs and emphasize verifying behavior in the browser after deploy. It fits indie operators who want a private control plane for coding agents rather than depending solely on hosted demos, and who need a structured credential worksheet instead of trial-and-error env copying.24installs8Remove Demo Limitsremove-demo-limits is an agent skill for maintainers who fork Vercel Open Agents / Open Harness templates and want full self-hosting instead of managed-template trial behavior. It walks through reading the trial module and the sessions, chat, and message-deletion API surfaces (and their tests) before editing, then removing guards that block session creation, chat messages, or deletes for trial users. The goal is a fork that behaves like a normal deployment: no session or message caps, no hosted-only restriction errors, and no demo-specific delete blocks. Solo builders shipping their own agent harness on their infra use it when phrases like remove demo limits or remove managed template restrictions appear. It is a focused codebase surgery skill, not a general deployment guide.24installs