AI coding¶
Most developers use an AI coding agent when they pick up a new API. SweatStack is built to make that path short. Three things help your agent ship working code instead of guessing.
1. Install the skills¶
SweatStack ships installable skills for AI coding agents. Your agent loads them on demand and writes code with the correct imports, real method signatures, and current authentication patterns. No more pasting docs into the chat window.
There are two skills:
| Skill | Covers | Install |
|---|---|---|
sweatstack |
The REST API, OAuth2 and OpenID Connect, Sign in with SweatStack, single-file web apps, the CLI | npx skills add SweatStack/sweatstack-skills |
sweatstack-python |
The Python SDK: clients, authentication, activities and longitudinal queries, dailies, tests, frames, Streamlit and FastAPI | npx skills add SweatStack/sweatstack-python |
The Python skill lives in the SDK's repository, so it describes the SDK version it ships with.
Supported agents. Anything that reads the Agent Skills format. That covers Claude Code, Claude Desktop, OpenAI Codex, Cursor and a growing list of agent frameworks.
2. Point your agent at the docs¶
If your agent can fetch URLs (most can), the next-best step after the skills is to point it at the docs directly.
| Resource | URL |
|---|---|
| These docs | https://docs.sweatstack.no/ |
| These docs as Markdown: an index of every page, for agents | https://docs.sweatstack.no/llms.txt |
| API reference (every endpoint and schema, in-site) | https://docs.sweatstack.no/learn/api-reference/ |
| API playground (live, auto-authenticated UI) | https://app.sweatstack.no/docs |
| Raw OpenAPI schema | https://app.sweatstack.no/openapi.json |
Useful one-liner to drop into your system prompt:
Before writing any SweatStack code, fetch
https://docs.sweatstack.no/getting-started/build/and the OpenAPI schema athttps://app.sweatstack.no/openapi.json.
3. Prompt for the right surface¶
Four patterns that produce working code on the first try:
- Default to the Python SDK. It's the most documented surface, and your agent produces a runnable script faster with it than with raw HTTP.
- Ask for a FastAPI snippet when you need OAuth. The
sweatstack.fastapihelper handles routes, sessions, and callbacks. The agent should use it instead of building OAuth from scratch. - Pin the SDK version if you care about reproducibility. Old training data uses method names from earlier releases, like
get_activities(). Calling a removed name raises an error that names the replacement, so an agent that runs its code fixes it. - Use the Build an app quickstart as a reference example. Telling your agent "follow the quickstart shape" gets you a clean starting point.
Every page has a Markdown version for agents: add index.md to its URL, like https://docs.sweatstack.no/learn/python/index.md. llms.txt lists them all.