Install any skill in seconds. Free to start, no credit card required.
Get Started Free →All public functions must validate arguments and return types
.claude/skills/kunanonj-cursor-plugin-convex-rule-argument-validation/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -6% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -42% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -42% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -21% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 36% | 0% |
All public query, mutation, and action functions MUST define validators for both arguments and return types. This protects against malicious input and provides type safety.
Always use the args and returns fields:
typescriptexport const createTask = mutation({ args: { text: v.string(), userId: v.id("users"), priority: v.optional(v.union( v.literal("low"), v.literal("medium"), v.literal("high") )), }, returns: v.id("tasks"), handler: async (ctx, args) => { return await ctx.db.insert("tasks", { text: args.text, userId: args.userId, priority: args.priority ?? "medium", completed: false, }); }, });
Internal functions (from internal.*) can skip validators if they're only called by trusted backend code, but it's still recommended.
Enable the @convex-dev/require-argument-validators ESLint rule to enforce this automatically.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 5,344 | 3,573 | -33% | 1 | 1 | 0% | 1,126 | 1,061 | -6% | 0 | 0 | — |
case-02 | fail→pass | 10,396 | 5,472 | -47% | 1 | 1 | 0% | 2,348 | 1,373 | -42% | 0 | 0 | — |
case-03 | fail→pass | 10,412 | 5,007 | -52% | 1 | 1 | 0% | 2,222 | 1,298 | -42% | 0 | 0 | — |
case-04 | fail→pass | 7,932 | 5,217 | -34% | 1 | 1 | 0% | 1,593 | 1,264 | -21% | 0 | 0 | — |
case-05 | fail→pass | 4,526 | 4,413 | -2% | 1 | 1 | 0% | 892 | 1,215 | +36% | 0 | 0 | — |
case-06 | fail→pass | 7,108 | 4,122 | -42% | 1 | 1 | 0% | 1,510 | 1,121 | -26% | 0 | 0 | — |
case-07 | fail→pass | 4,263 | 3,011 | -29% | 1 | 1 | 0% | 852 | 854 | +0% | 0 | 0 | — |
case-20 | pass→pass | 9,164 | 5,276 | -42% | 1 | 1 | 0% | 1,662 | 1,343 | -19% | 0 | 0 | — |
case-08 | fail→pass | 3,340 | 2,862 | -14% | 1 | 1 | 0% | 547 | 756 | +38% | 0 | 0 | — |
case-09 | fail→pass | 5,536 | 3,925 | -29% | 1 | 1 | 0% | 1,186 | 1,009 | -15% | 0 | 0 | — |
case-10 | fail→pass | 13,126 | 1,892 | -86% | 1 | 1 | 0% | 2,613 | 503 | -81% | 0 | 0 | — |
case-11 | fail→pass | 5,058 | 3,958 | -22% | 1 | 1 | 0% | 943 | 1,089 | +15% | 0 | 0 | — |
case-12 | fail→pass | 5,875 | 4,829 | -18% | 1 | 1 | 0% | 1,161 | 1,273 | +10% | 0 | 0 | — |
case-13 | fail→pass | 8,554 | 5,121 | -40% | 1 | 1 | 0% | 1,767 | 1,257 | -29% | 0 | 0 | — |
case-14 | fail→pass | 4,945 | 4,067 | -18% | 1 | 1 | 0% | 1,069 | 1,125 | +5% | 0 | 0 | — |
case-15 | fail→pass | 5,093 | 4,628 | -9% | 1 | 1 | 0% | 1,015 | 1,199 | +18% | 0 | 0 | — |
case-16 | fail→pass | 6,134 | 3,553 | -42% | 1 | 1 | 0% | 1,211 | 944 | -22% | 0 | 0 | — |
case-17 | fail→pass | 6,031 | 2,420 | -60% | 1 | 1 | 0% | 1,110 | 732 | -34% | 0 | 0 | — |
case-18 | fail→pass | 4,069 | 3,122 | -23% | 1 | 1 | 0% | 809 | 979 | +21% | 0 | 0 | — |
case-19 | fail→pass | 3,839 | 6,004 | +56% | 1 | 1 | 0% | 752 | 1,478 | +97% | 0 | 0 | — |
case-21 | pass→pass | 5,601 | 4,418 | -21% | 1 | 1 | 0% | 1,183 | 1,063 | -10% | 0 | 0 | — |
case-22 | pass→pass | 9,286 | 4,926 | -47% | 1 | 1 | 0% | 1,767 | 1,352 | -23% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +86 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.