Install any skill in seconds. Free to start, no credit card required.
Get Started Free →HTTP outcalls performed by the backend canister (not in the frontend).
.claude/skills/aiskillstore-extension-http-outcalls/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -44% | 0% |
| case-01 | ✗→✓ | ▲ Improved | -54% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -52% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -46% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -53% | 0% |
HTTP outcalls extension for Caffeine AI.
This skill adds the ability to make HTTP GET and POST requests from the backend canister. Useful for integrating with external APIs and services.
For HTTP outcalls that must be performed in the backend:
There is the prefabricated module mo:caffeineai-http-outcalls/outcall.mo that that cannot be modified. It provides fundamental functionality for making HTTP GET or PUT requests in the backend.
mo:caffeineai-http-outcalls/outcallmodule { public type TransformationInput = { context : Blob; response : IC.HttpRequestResult; }; public type TransformationOutput = IC.HttpRequestResult; public type Transform = query TransformationInput -> async TransformationOutput; public type Header = { name: Text; value: Text; }; // Helper function for the transform callback used by the IC on HTTP outcalls. public func transform(input : TransformationInput) : TransformationOutput; // HTTP GET request with a transform callback function. public func httpGetRequest(url : Text, extraHeaders: [Header], transform : Transform) : async Text; // HTTP POST request, specifying a transform callback. public func httpPostRequest(url : Text, extraHeaders: [Header], body : Text, transform : Transform) : async Text; };
Usage for GET:
motoko filepath=src/backend/main.moimport OutCall "mo:caffeineai-http-outcalls/outcall"; actor { public query func transform(input: OutCall.TransformationInput) : async OutCall.TransformationOutput { OutCall.transform(input); }; func makeGetOutcall(url: Text) : async Text { await OutCall.httpGetRequest(url, [], transform); }; };
Hint: JSON parsing is not directly supported in Motoko. Better tunnel JSON to frontend for parsing.
POST usage is analogous.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-04 | fail→pass | 28,629 | 11,824 | -59% | 1 | 1 | 0% | 3,363 | 1,877 | -44% | 0 | 0 | — |
case-01 | fail→pass | 40,975 | 10,607 | -74% | 1 | 1 | 0% | 3,339 | 1,520 | -54% | 0 | 0 | — |
case-02 | fail→pass | 23,752 | 11,519 | -52% | 1 | 1 | 0% | 3,517 | 1,704 | -52% | 0 | 0 | — |
case-03 | fail→pass | 25,291 | 13,302 | -47% | 1 | 1 | 0% | 3,776 | 2,050 | -46% | 0 | 0 | — |
case-05 | fail→pass | 19,033 | 9,181 | -52% | 1 | 1 | 0% | 2,529 | 1,189 | -53% | 0 | 0 | — |
case-06 | fail→pass | 17,446 | 8,298 | -52% | 1 | 1 | 0% | 2,209 | 1,110 | -50% | 0 | 0 | — |
case-07 | pass→pass | 17,368 | 10,134 | -42% | 1 | 1 | 0% | 1,958 | 1,440 | -26% | 0 | 0 | — |
case-08 | fail→pass | 13,711 | 8,350 | -39% | 1 | 1 | 0% | 1,490 | 1,113 | -25% | 0 | 0 | — |
case-09 | pass→pass | 14,893 | 11,004 | -26% | 1 | 1 | 0% | 1,483 | 1,425 | -4% | 0 | 0 | — |
case-10 | fail→pass | 27,718 | 7,948 | -71% | 1 | 1 | 0% | 1,407 | 999 | -29% | 0 | 0 | — |
case-11 | fail→pass | 13,927 | 7,896 | -43% | 1 | 1 | 0% | 1,397 | 939 | -33% | 0 | 0 | — |
case-12 | pass→pass | 12,591 | 13,735 | +9% | 1 | 1 | 0% | 2,549 | 2,078 | -18% | 0 | 0 | — |
case-13 | pass→pass | 10,998 | 14,919 | +36% | 1 | 1 | 0% | 2,014 | 2,237 | +11% | 0 | 0 | — |
case-14 | pass→pass | 14,155 | 9,976 | -30% | 1 | 1 | 0% | 1,742 | 1,333 | -23% | 0 | 0 | — |
case-15 | pass→pass | 9,296 | 2,729 | -71% | 1 | 1 | 0% | 1,543 | 958 | -38% | 0 | 0 | — |
case-16 | fail→pass | 16,995 | 4,740 | -72% | 1 | 1 | 0% | 2,115 | 1,405 | -34% | 0 | 0 | — |
case-17 | fail→pass | 26,663 | 7,230 | -73% | 1 | 1 | 0% | 2,856 | 847 | -70% | 0 | 0 | — |
case-18 | fail→pass | 15,703 | 2,405 | -85% | 1 | 1 | 0% | 1,925 | 821 | -57% | 0 | 0 | — |
case-19 | fail→pass | 13,624 | 4,707 | -65% | 1 | 1 | 0% | 1,378 | 1,016 | -26% | 0 | 0 | — |
case-20 | fail→pass | 8,409 | 3,191 | -62% | 1 | 1 | 0% | 1,360 | 1,014 | -25% | 0 | 0 | — |
case-21 | pass→pass | 19,133 | 13,182 | -31% | 1 | 1 | 0% | 2,655 | 2,080 | -22% | 0 | 0 | — |
case-22 | fail→pass | 10,867 | 9,447 | -13% | 1 | 1 | 0% | 1,911 | 1,421 | -26% | 0 | 0 | — |
case-23 | fail→fail | 12,319 | 4,855 | -61% | 1 | 1 | 0% | 1,967 | 1,275 | -35% | 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. 23 cases were attempted. The headline lift of +65 percentage points is the difference between those two pass rates over the 23 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.