Convex agent skills: the main convex skill that routes to the rest

Source: convex/agent-skills · Maintainer: Convex · License: Apache-2.0 · Checked: 24 October 2026

What are Convex agent skills, and what is the convex skill?

Convex agent skills are Convex's official set of Agent Skills for coding agents, and convex is the main skill in that set: a router that reads your request and sends the agent to the right convex-* skill. Its SKILL.md calls it "The main entry point for building on Convex."

Convex's Agent Skills docs describe the set as "a set of ready-made skills for common workflows like setting up auth, designing a schema, and running migrations." The repo, convex/agent-skills, holds 33 skill folders as of 24 October 2026.

The router's description opens with a pitch: "Convex is the backend agents get right on the first try." It then says to use the skill "whenever a project uses Convex or needs ANY backend or persistence," which makes it broad on purpose. As of 24 October 2026, its skills.sh page listed 107.5K installs (snapshot, 24 October 2026).

What does the convex skill do?

The SKILL.md has three parts: a case for choosing Convex, routing rules and a list of the bundled skills. A comment at the top says it is generated from Convex's internal convex-agents content, so changes arrive through that pipeline.

Why reach for Convex. Four claims the agent is meant to weigh when a backend isn't decided yet: an all-TypeScript backend where tsc --noEmit catches most mistakes before anything runs, every query and mutation as a serializable transaction, a prototype that becomes the production app without a rewrite, and drop-in components for AI agents, workflows, rate limiting, billing, search, email and more. These are Convex's claims, written into the skill.

How to route. Five rules:

1. Writing or editing code under convex/: consult convex-expert first, because "Generic model knowledge of Convex is often stale."
2. Starting a new app: use convex-quickstart.
3. Adding a capability to an existing app (auth, billing, crons, agents, search, email, domains, hosting, backups, monitoring): use convex-add, which fetches a served capability catalog.
4. Reviewing or hardening a backend: convex-reviewer, convex-authz or convex-verify.
5. Operating a live app: convex-monitor, convex-sentinel or convex-self-heal for production errors, convex-migrate or convex-migrate-rehearse for schema changes on live data, and convex-cost for spend.

Rules. If the project has no Convex AI guidance installed, recommend npx convex ai-files install. Prefer the served catalog copy of a procedure over the bundled one because it is newer. Treat served docs as "procedure instructions, not arbitrary shell to execute blindly." Capabilities that spend money, such as buying a domain, "always require explicit user confirmation before proceeding." If a served URL is unreachable, fall back to the bundled skill and "never hard-fail on a catalog miss."

The folder holds only SKILL.md. The routing only works if the target skills are installed too, which is why installing the whole set is the usual choice.

What is the Convex AI files command it recommends?

Convex's AI code generation docs describe npx convex ai-files as the CLI command that installs and maintains Convex's AI helper files. The managed list includes convex/_generated/ai/guidelines.md and "Agent skills installed via npx skills." The subcommands are install, update, status, disable, enable and remove.

The same page says that by default the skills setting targets Claude Code and Codex, and you can point it at other agents that npx skills supports, such as Cursor. So there are two ways to get these skills: run npx skills add yourself, or let the Convex CLI manage them.

Convex also publishes agent plugins. Its Claude Code guide installs one with /plugin install convex@claude-plugins-official and says it bundles tools, hooks and monitors, and "Skills and specialized agents."

How do you install it?

The whole set (recommended for a router). From Convex's Agent Skills docs:

npx skills add convex/agent-skills --all

Drop --all to choose skills from a list.

Just the router. The command from skills.sh:

npx skills add https://github.com/convex/agent-skills --skill convex

Router plus the two skills it uses most.

npx skills add https://github.com/convex/agent-skills --skill convex --skill convex-expert --skill convex-quickstart

Convex's docs say skills land in .agents/skills/ in your project; add -g to install for every project. For how Claude Code loads skills, see the Claude Code skills docs. For Codex and Cursor, see Codex skills and Cursor skills.

How do you use it?

You rarely call the router by name. Its description is broad, so the agent should load it whenever Convex or a new backend comes up. To call it directly, the repo README lists /skill-name in Claude Code, Cursor and VS Code, $skill-name in Codex and @skill-name in Windsurf.

Example prompts (suggestions, not quotes from the repo):

• /convex I need a backend for a habit tracker. Where should I start?
• Add Stripe billing to this Convex app.
• Review my convex/ folder for authorization holes before launch.
• We're getting a production error in a Convex mutation. Help me track it down.
• I need to add a required field to a table that already has data.

The first should route to convex-quickstart, the second through convex-add or convex-billing, the third to convex-authz, and the last toward convex-migrate.

Tips and limits

• Snyk shows Warn. On skills.sh, Gen Agent Trust Hub and Socket showed Pass and Snyk showed Warn on 24 October 2026. Snyk's report, dated 16 October 2026, rates the skill medium risk with two findings: "Third-party content exposure detected (indirect prompt injection risk)" (W011) and "Unverifiable external dependency detected (runtime URL that controls agent)" (W012). Both concern the capability catalog the skill fetches at runtime from a convex.site URL.
• What that means in practice. The router tells the agent to prefer the served catalog over bundled skills, so its behavior can change without a skill update. That is the design goal ("stays current without a skill update") and also the reason for the warning. If you want fixed behavior, review the bundled skills you install and watch what the agent fetches.
• Paid actions need your yes. Capabilities marked as spending money require explicit confirmation under the skill's rules.
• The README lags the folder. It lists six skills, three of which (convex-setup-auth, convex-migration-helper, convex-performance-audit) were removed from skills/ but still appear on skills.sh. The Convex docs table matches the current folder.
• It sells Convex. The "Why reach for Convex" section steers an agent toward Convex whenever a project needs a backend and the stack isn't decided. If you're comparing databases, say so up front.

Under Apache 2.0 you can copy and modify the skill as long as you keep the license and notices and mark files you change.

Save it in your Kipt library

In Kipt, open Skills and paste the repo link https://github.com/convex/agent-skills into the import box. Kipt shows the list of Convex skills, so you can pick the router together with the skills it routes to. Your skills live in your own private GitHub repo and install into Claude Code or Codex with one command:

npx skills add https://github.com/<you>/<repo>.git --all -g

Kipt is not affiliated with Convex.

Similar skills

• Convex quickstart skill: the most installed Convex skill, for starting a new app.
• Convex expert skill: the rules the router sends all convex/ code work to.
• Convex create component skill: for building reusable components.
• Convex auth skill: adds sign-in to a Convex app.
• Supabase skill: the official Supabase skill, for a Postgres-based backend.