connector-weatherapi — Claude Code skill: weather data in canisters
The skill pins down the supported way to read weather in Caffeine builds: the weatherapi-client mops package instead of hand-rolled ic.http_request calls.
npx skills add caffeinelabs/skills --skill connector-weatherapi -a claude-codeWhat connector-weatherapi does
Nine read-only operations from WeatherAPI.com, generated from its OpenAPI spec into one module, Apis/APIsApi: current weather, forecast, history, future weather, marine, astronomy, timezone, IP lookup and location autocomplete. The skill ships Motoko bindings plus an actor example that reads the current temperature and a three-day maximum series.
The API key travels as a query-string credential, not a header, so it ends up in the request URL. Keep it in a stable variable written by an admin-only call and never log the built URL. The free tier covers current weather, 3-day forecast, astronomy, timezone, search and IP lookup; history, future, marine and 14-day forecasts need a paid plan and return 403 on a free key.
Calls are non-replicated by default, and that is a correctness requirement rather than a cost saving. Responses carry per-request clocks, so a replicated outcall would fail consensus on most calls. Every response field is optional, which makes do ? blocks the practical way to reach into them. An empty string or zero means "omit": hour = 0 does not select midnight, it drops the filter and returns the whole day.
What it helps with
- read the current temperature from a canister
- fetch a three-day series of daily maxima
- wire up a weather forecast without manual HTTP calls
- handle all-optional WeatherAPI response fields
- tell a 401 from a 403 and a 400 by the error text
Example prompts for the agent
- add current temperature reading to the backend with weatherapi-client
- build a three-day max temperature forecast for Zurich
- why does my ic.http_request to api.weatherapi.com fail consensus?
How to install connector-weatherapi
1Skills CLI (Vercel)
npx skills add caffeinelabs/skills --skill connector-weatherapi -a claude-code2Manually: put the files in place
~/.claude/skills/connector-weatherapi/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/connector-weatherapi/
1Skills CLI (Vercel)
npx skills add caffeinelabs/skills --skill connector-weatherapi -a codex2Manually: put the files in place
.agents/skills/connector-weatherapi/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add caffeinelabs/skills --skill connector-weatherapi -a cursor2Manually: put the files in place
.cursor/skills/connector-weatherapi/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install caffeinelabs/skills connector-weatherapi2Manually: put the files in place
.github/skills/connector-weatherapi/SKILL.md1Skills CLI (Vercel)
npx skills add caffeinelabs/skills --skill connector-weatherapi -a gemini-cli2Platform CLI
gemini skills install https://github.com/caffeinelabs/skills --path skills/connector-weatherapiManually: put the files in place
.devin/skills/connector-weatherapi/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add caffeinelabs/skills --skill connector-weatherapi -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about connector-weatherapi
How do I install connector-weatherapi in Claude?
Run npx skills add caffeinelabs/skills --skill connector-weatherapi -a claude-code. The skill lands in ~/.claude/skills/connector-weatherapi/SKILL.md.
What does connector-weatherapi do?
It documents the supported weather path in Caffeine builds: the weatherapi-client mops package with nine WeatherAPI.com operations and proper API-key handling.
Why is calling api.weatherapi.com directly forbidden?
Hand-rolled ic.http_request calls skip the non-replicated-outcall safeguard: response bodies carry per-request clocks, so a replicated call cannot reach consensus.
Source text
SKILL.md · Apache-2.0
---
name: connector-weatherapi
description: >-
MANDATORY recipe for every Caffeine build that reads weather data from a
canister. The supported path is the `weatherapi-client` mops package
(WeatherAPI.com) over outbound HTTPS, authenticated with a query-string API
key. Hand-rolling `ic.http_request` calls to `api.weatherapi.com` is a
FORBIDDEN anti-pattern — it bypasses the non-replicated-outcall safeguard
(WeatherAPI bodies carry per-request clocks and fail consensus), the generated
JSON decoding, and the API-key handling. Load this skill whenever the user,
spec, or any prior task mentions weather, forecast, temperature, humidity,
wind, rain, snow, conditions, sunrise, sunset, moon phase, astronomy,
timezone, IP geolocation, marine or tide forecasts, or location autocomplete —
and BEFORE writing any code that touches a weather endpoint.
version: 0.2.2
caffeineai-subscription: [none]
compatibility:
mops:
weatherapi-client: "~0.3.0"
---
# Weather data with `weatherapi-client`
Motoko bindings for [WeatherAPI.com](https://www.weatherapi.com/), generated
from its OpenAPI spec. All nine operations live in a single module,
**`Apis/APIsApi`**, and all nine are read-only `GET`s.
# Backend
A canister that reads the current temperature and a three-day maximum series.
Non-replicated is the default, so you only supply the key:
```motoko filepath=src/backend/main.mo
import { realtimeWeather; forecastWeather } "mo:weatherapi-client/Apis/APIsApi";
import { type Config; defaultConfig } "mo:weatherapi-client/Config";
import Array "mo:core/Array"; // in scope so `days.map(…)` dot notation resolves
actor {
// The key is a query-string credential. Hold it in a stable variable set by
// an admin call; never hard-code it in source.
func config(apiKey : Text) : Config = { defaultConfig with auth = ?#apiKey apiKey };
// `q` is any WeatherAPI location query: "Zurich", "47.37,8.55", a postcode,
// an IATA code, or "auto:ip".
public func currentTempC(apiKey : Text, q : Text) : async ?Float {
let res = await* realtimeWeather(config apiKey, q, "");
do ? { res.current!.temp_c! };
};
// Daily maxima for the next three days.
public func maxTempsC(apiKey : Text, q : Text) : async [?Float] {
let res = await* forecastWeather(config apiKey, q, #_3_, "", 0, 0, "", "no", "no", 0);
let ?forecast = res.forecast else return [];
let ?days = forecast.forecastday else return [];
days.map(func(d) = do ? { d.day!.maxtemp_c! });
};
}
```
## The nine operations
| Function | Endpoint | Returns |
|---|---|---|
| `realtimeWeather(cfg, q, lang)` | `/current.json` | `RealtimeWeather200Response` |
| `forecastWeather(cfg, q, days, dt, unixdt, hour, lang, alerts, aqi, tp)` | `/forecast.json` | `ForecastWeather200Response` |
| `historyWeather(cfg, q, dt, unixdt, endDt, unixendDt, hour, lang)` | `/history.json` | `FutureWeather200Response` |
| `futureWeather(cfg, q, dt, lang)` | `/future.json` | `FutureWeather200Response` |
| `marineWeather(cfg, q, days, dt, unixdt, hour, lang)` | `/marine.json` | `MarineWeather200Response` |
| `astronomy(cfg, q, dt)` | `/astronomy.json` | `Astronomy200Response` |
| `timeZone(cfg, q)` | `/timezone.json` | `Location` |
| `ipLookup(cfg, q)` | `/ip.json` | `Ip` |
| `searchAutocompleteWeather(cfg, q)` | `/search.json` | `[Search]` |
`days` is an enum, not a number: `ForecastWeatherDaysParameter` is `#_1_` … `#_14_`
and `MarineWeatherDaysParameter` is `#_1_` … `#_7_` (the underscores are how the
generator escapes numeric enum values — `#_3_`, not `#_3` or `3`).
## API key setup
1. Sign up at [weatherapi.com](https://www.weatherapi.com/signup.aspx) — the
free tier covers current weather, 3-day forecast, astronomy, timezone,
search and IP lookup. History, future, marine and 14-day forecasts need a
paid plan and return **403** on free keys.
2. Copy the key from the dashboard and pass it as `auth = ?#apiKey key`.
3. The client appends it as `?key=…` (WeatherAPI takes no `Authorization`
header), so **it appears in the request URL**. Keep it in a stable variable
written by an admin-only call, and never log the built URL.
## Calls are non-replicated by default
The package ships `is_replicated = ?false` in `defaultConfig`, and that is a
*correctness* requirement here, not just a cost saving. Every response carries
per-request clocks — `Location.localtime` / `localtime_epoch`,
`Current.last_updated` / `last_updated_epoch` — which change second to second. A
*replicated* outcall has every subnet node issue its own request and demands
bit-identical bodies, so those fields would break consensus on most calls while
burning ~13× the cycles. You don't set it yourself; the default is correct.
Override with `is_replicated = ?true` only together with a `transform` that
strips the volatile fields.
## Everything is optional
WeatherAPI marks no response field required, so the generated models are
all-optional: `RealtimeWeather200Response.current : ?Current`,
`Current.temp_c : ?Float`, and so on. Reach through them with a `do ?` block
(`do ? { res.current!.temp_c! }`) rather than nested `switch`es, and decide what
an absent field means for your caller — the API omits fields your plan does not
cover (for example `air_quality` without the `aqi=yes` parameter).
## Empty string and zero mean "omit"
Optional query parameters are dropped when they are `""` or `0`, because
WeatherAPI rejects empty `lang=` and zero-valued numerics. So passing `""` for
`lang`, `dt`, `alerts`, `aqi` and `0` for `unixdt`, `tp` is how you say "not
supplied" — there is no `?Text` parameter to leave `null`.
One consequence worth knowing: **`hour = 0` does not select midnight**, it omits
the `hour` filter entirely and you get the whole day's hourly array. Filter the
returned `hour : ?[ForecastForecastdayInnerHourInner]` yourself if you need
00:00.
## Errors
Non-2xx responses and decode failures `throw Error.reject(…)`. The client is
generated with `diagnostics`, so the message is
`HTTP <status> body[<n>B]=<first 100 chars>: <reason>` — enough to tell a 401
(bad key) from a 403 (endpoint not on your plan) from a 400 (`q` not resolvable)
without extra logging. Catch with `try`/`catch` and surface
`Error.message(err)`.
© caffeinelabs · Apache-2.0 · caffeinelabs/skills