Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Only schedule internal functions, never api functions
.claude/skills/kunanonj-cursor-plugin-convex-rule-scheduler-usage/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 26% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -32% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -44% | 0% |
| case-21 | ✓→✗ | ▼ Worse | -15% | 0% |
| case-05 | ✓→✓ | = Same ✓ | -24% | 0% |
Always schedule internal functions with ctx.scheduler, ctx.runAfter, ctx.runAt, and ctx.run* methods. Never schedule public api functions.
Scheduled functions bypass authentication and argument validation that public APIs expect to receive from clients.
Bad:
typescriptimport { api } from "./_generated/api"; export const processPayment = action({ handler: async (ctx, args) => { // ❌ Don't schedule api functions await ctx.scheduler.runAfter(0, api.users.chargeUser, { userId: args.userId, amount: args.amount, }); }, });
Good:
typescriptimport { internal } from "./_generated/api"; // Define an internal function export const chargeUserInternal = internalMutation({ args: { userId: v.id("users"), amount: v.number() }, handler: async (ctx, args) => { // ... charging logic ... }, }); export const processPayment = action({ handler: async (ctx, args) => { // ✅ Schedule internal functions await ctx.scheduler.runAfter(0, internal.users.chargeUserInternal, { userId: args.userId, amount: args.amount, }); }, });
Use internalQuery, internalMutation, and internalAction for functions that should only be called from backend code:
typescriptexport const internalHelper = internalMutation({ args: { /* ... */ }, handler: async (ctx, args) => { // No auth check needed - only callable from backend }, });
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-05 | pass→pass | 9,298 | 5,510 | -41% | 1 | 1 | 0% | 2,137 | 1,629 | -24% | 0 | 0 | — |
case-06 | pass→pass | 8,815 | 5,696 | -35% | 1 | 1 | 0% | 2,134 | 1,622 | -24% | 0 | 0 | — |
case-01 | fail→pass | 4,688 | 4,233 | -10% | 1 | 1 | 0% | 1,086 | 1,363 | +26% | 0 | 0 | — |
case-02 | fail→pass | 10,827 | 4,636 | -57% | 1 | 1 | 0% | 2,195 | 1,493 | -32% | 0 | 0 | — |
case-03 | pass→pass | 10,157 | 6,203 | -39% | 1 | 1 | 0% | 2,362 | 1,815 | -23% | 0 | 0 | — |
case-04 | fail→pass | 13,863 | 4,814 | -65% | 1 | 1 | 0% | 2,667 | 1,482 | -44% | 0 | 0 | — |
case-07 | pass→pass | 8,406 | 5,055 | -40% | 1 | 1 | 0% | 1,790 | 1,634 | -9% | 0 | 0 | — |
case-08 | pass→pass | 10,608 | 8,066 | -24% | 1 | 1 | 0% | 2,556 | 2,368 | -7% | 0 | 0 | — |
case-09 | pass→pass | 13,851 | 6,118 | -56% | 1 | 1 | 0% | 3,177 | 1,818 | -43% | 0 | 0 | — |
case-10 | pass→pass | 8,011 | 5,658 | -29% | 1 | 1 | 0% | 1,859 | 1,864 | +0% | 0 | 0 | — |
case-11 | pass→pass | 11,695 | 6,835 | -42% | 1 | 1 | 0% | 2,791 | 2,136 | -23% | 0 | 0 | — |
case-12 | pass→pass | 7,066 | 5,469 | -23% | 1 | 1 | 0% | 1,499 | 1,782 | +19% | 0 | 0 | — |
case-13 | pass→pass | 10,764 | 6,071 | -44% | 1 | 1 | 0% | 2,439 | 1,624 | -33% | 0 | 0 | — |
case-14 | pass→pass | 12,085 | 7,234 | -40% | 1 | 1 | 0% | 2,601 | 1,969 | -24% | 0 | 0 | — |
case-15 | pass→pass | 8,331 | 5,890 | -29% | 1 | 1 | 0% | 2,103 | 1,821 | -13% | 0 | 0 | — |
case-20 | pass→pass | 6,801 | 4,356 | -36% | 1 | 1 | 0% | 1,533 | 1,303 | -15% | 0 | 0 | — |
case-16 | pass→pass | 12,856 | 10,806 | -16% | 1 | 1 | 0% | 2,758 | 3,027 | +10% | 0 | 0 | — |
case-17 | pass→pass | 12,235 | 5,315 | -57% | 1 | 1 | 0% | 2,808 | 1,473 | -48% | 0 | 0 | — |
case-18 | pass→pass | 7,191 | 5,003 | -30% | 1 | 1 | 0% | 1,545 | 1,539 | -0% | 0 | 0 | — |
case-19 | fail→fail | 9,594 | 4,886 | -49% | 1 | 1 | 0% | 1,799 | 1,502 | -17% | 0 | 0 | — |
case-21 | pass→fail | 11,438 | 4,941 | -57% | 1 | 1 | 0% | 1,799 | 1,536 | -15% | 0 | 0 | — |
case-22 | pass→pass | 8,408 | 6,387 | -24% | 1 | 1 | 0% | 1,582 | 1,584 | +0% | 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 +9 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
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.