Convex Development General

Convex Development General Skill

Schema and Validators

• Always define table schemas using defineTable(v.object({...})) in convex/schema.ts.
• Use v.id("tableName") for cross-document references — never plain v.string().
• Omit _id and _creationTime from schema definitions — they are auto-generated system fields.
• See https://docs.convex.dev/database/types for all available validator types.

Function Registration

• Use new function syntax: query({ args: {}, returns: v.null(), handler: async (ctx, args) => {...} }).
• ALWAYS include both argument (args) and return (returns) validators; if nothing is returned, use returns: v.null().
• Use internalQuery/internalMutation/internalAction for private functions — never expose internal logic via public API.
• Use httpAction with httpRouter for HTTP endpoints in convex/http.ts.

Index-First Query Patterns

• Prefer .withIndex("by_field", (q) => q.eq("field", value)) over .filter((q) => q.eq(q.field("field"), value)).
• Add indexes to schema.ts using .index("name", ["field1", "field2"]) on defineTable.
• Use .withSearchIndex for full-text search patterns.
• Avoid full table scans with .collect() on large tables — use .paginate(opts) or .take(n).

Queries vs Mutations vs Actions

• query: read-only, reactive (subscriptions), runs in V8 sandbox.
• mutation: database writes, transactional, runs in V8 sandbox.
• action: can call external APIs / run Node.js, NOT transactional — minimize direct db access.
• Use ctx.runQuery/ctx.runMutation for cross-function calls; avoid action-to-mutation loops that split transactions.

Await All Promises

• Always await ctx.db.patch(...), await ctx.scheduler.runAfter(...), etc.
• Enable no-floating-promises ESLint rule to catch un-awaited Convex calls.

Real-Time Subscriptions

• Client-side useQuery hooks auto-subscribe and re-render on data changes — no manual onSnapshot wiring needed.
• Keep query functions deterministic to maximize cache hit rate.

export default defineSchema({ messages: defineTable({ channel: v.id("channels"), // cross-doc ref body: v.string(), user: v.id("users"), // _id and _creationTime are auto-added — do NOT include them }) .index("by_channel", ["channel"]) .index("by_channel_user", ["channel", "user"]), });

// convex/messages.ts — correct query with index + return validator import { query } from "./_generated/server"; import { v } from "convex/values";

export const getByChannel = query({ args: { channelId: v.id("channels") }, returns: v.array(v.object({ _id: v.id("messages"), body: v.string() })), handler: async (ctx, args) => { return await ctx.db .query("messages") .withIndex("by_channel", (q) => q.eq("channel", args.channelId)) .take(50); // bounded — never unbounded .collect() in production }, });

// convex/messages.ts — internal mutation (not exposed publicly) import { internalMutation } from "./_generated/server";

export const deleteOldMessages = internalMutation({ args: { before: v.number() }, returns: v.null(), handler: async (ctx, args) => { const old = await ctx.db .query("messages") .withIndex("by_channel", (q) => q.lt("_creationTime", args.before)) .take(100); await Promise.all(old.map((msg) => ctx.db.delete(msg._id))); }, });

</examples>

## Iron Laws

1. **ALWAYS** define document schemas using Convex `v` validators — never rely on raw TypeScript types alone for runtime-enforced schema correctness.
2. **NEVER** manually include `_id` or `_creationTime` fields in schema definitions — they are automatically generated system fields and specifying them causes runtime errors.
3. **ALWAYS** use `v.id("tableName")` for cross-document references — never store foreign keys as plain strings, which bypasses Convex's referential integrity tools.
4. **NEVER** perform direct database mutations from client-side code — all mutations must be defined as Convex mutation functions in the `convex/` directory.
5. **ALWAYS** add `.withIndex(...)` for filtered queries on non-trivial tables — never use `.filter()` as a substitute for a missing index on production data, and never use `.collect()` without a bound (`take(n)` or `.paginate()`) on large tables.

## Anti-Patterns

| Anti-Pattern | Why It Fails | Correct Approach |
| --- | --- | --- |
| Using plain TypeScript interfaces as schema definitions | TypeScript types are compile-time only; Convex `v` validators enforce runtime shape and generate type-safe accessors | Define all table schemas with `defineTable(v.object({...}))` |
| Adding `_id` or `_creationTime` to defineTable schemas | Convex rejects schemas that include system fields, causing runtime initialization errors | Omit system fields; access them via `doc._id` and `doc._creationTime` after query |
| Storing cross-document references as plain `v.string()` | Loses Convex's cross-reference validation and type inference for joined queries | Use `v.id("tableName")` so Convex validates the reference type |
| Running `.collect()` on large tables without pagination | Returns all documents, causing memory spikes and timeouts on large datasets | Use `.paginate(opts)` or `.take(100)` with cursor-based pagination |
| Writing to the database from React client code directly | Bypasses access control, validation, and audit trail; creates untraceable mutations | All writes must go through a Convex `mutation` function in `convex/` |
| Using `.filter()` instead of `.withIndex()` for field-based lookups | `.filter()` performs a full table scan; identical performance to filtering in-code but misses index speed-up | Define a schema index and use `.withIndex(name, q => q.eq(...))` |

## Memory Protocol (MANDATORY)

**Before starting:**

```bash
cat .claude/context/memory/learnings.md

After completing: Record any new patterns or exceptions discovered.

ASSUME INTERRUPTION: Your context may reset. If it's not in memory, it didn't happen.