The master skill in PlanetScale's official pack: runs a full read-only best-practices assessment of a database or org, then produces one evidence-backed report.
ConvexSkill
Summary
Convex's official top-level agent skill — routes an agent to the right convex-* skill for the task and to a served capability catalogue that stays current without a reinstall.
Features
- Routes to the right convex-* skill instead of loading every procedure at once
- Fetches a served capability catalogue so new capabilities need no reinstall
- Prefers the served procedure over the bundled copy, with a bundled fallback
- Points agents at convex-expert first for code under convex/
- Requires explicit confirmation for capabilities that spend money
- Works in Claude Code, Cursor, VS Code, Windsurf and Codex
Install This Skill
Add this skill to your favorite AI agent in a few steps.
Skill Content
Usage Instructions
Learn how to use this skill with different AI agents.
Example Usage
I'm starting a full-stack app and need a backend with auth and Stripe billing — figure out which Convex skills to use and set it up.
Description
Convex ships one skill per capability — more than thirty of them, all named convex-* — and this is the one that decides which of them a task needs. Install it and an agent working in a project with a convex/ directory stops guessing: it is told to consult convex-expert before writing backend code, convex-quickstart to scaffold a new app, convex-add to bolt on a capability, convex-reviewer / convex-authz / convex-verify to harden an existing one, and convex-monitor, convex-sentinel, convex-self-heal, convex-migrate or convex-cost when the app is already live.
Why a router skill is the right shape here. The alternative — one enormous Convex skill — puts every procedure in context for every task and goes stale the moment Convex ships a feature. This one fetches a served capability catalogue at runtime (capabilities.json, then the matched capability's own served markdown), so new capabilities become available to the agent without anyone reinstalling a skill. The skill's own rule is explicit: when a bundled procedure and a served procedure both exist, prefer the served copy because it is newer, and fall back to the bundled one only if the URL is unreachable.
The argument it makes for Convex itself is worth reading even if you have not decided on a backend, because it is really an argument about what agents get wrong. The whole backend is TypeScript with end-to-end types — schema, validators and function signatures flow to the client — so tsc --noEmit catches most generated-code mistakes before anything runs. Every query and mutation is automatically a serializable transaction, which means the race conditions and partial-write bugs that plague hand-wired backends cannot be expressed. That is a much better fit for code an agent wrote than a stack where correctness only shows up at runtime.
Guardrails baked in. Served documentation text is treated as procedure instructions, not as shell to execute blindly. Capabilities marked tier>0 — the ones that spend money, such as buying a domain — always require explicit user confirmation. And if the project has no Convex AI guidance installed, the skill recommends npx convex ai-files install first.
Install the whole set with npx skills add get-convex/agent-skills --all, or pick individual skills with the same command minus --all. Works in Claude Code, Cursor, VS Code with Copilot, Windsurf and Codex, which each have their own invocation prefix (/, @ or $).
Related Skills
Build crash-safe AI agents on AgentKit and step.ai — durable tool calls, human-in-the-loop approval, realtime progress, and when not to reach for an agent loop at all.
Inngest's durable-execution playbook as an agent skill — steps and memoisation, event and cron triggers, idempotency, cancellation, retries and non-retriable errors.
Temporal's official skill for building durable workflows — SDK patterns across seven languages, plus the determinism rules that decide whether a workflow survives a replay.