@nifrajs/runner

Run requests through a nifra app and capture structured results - the shared engine behind the website playground (humans) and the agent run/verify tool (an AI writes code → runs it → sees the failure → fixes it). It only touches app.fetch(Request) → Response, which is Web-standard, so it runs unchanged in the browser, Bun, Node, Deno, and on the edge - and depends on nothing.

import { runApp } from "@nifrajs/runner"
import { app } from "./app"

const results = await runApp(app, [
  { path: "/users/1" },
  { method: "POST", path: "/users", body: { name: "Ada" } },
])
// → [{ method, path, ok, status, headers, body, durationMs }, …]

• runApp(app, requests, options?) - run a batch in order; one RunResult per request. Never throws: a crash on any request becomes that result's error and the batch continues.
• runRequest(app, spec, options?) - a single request.

A plain object/array body is JSON-encoded (with a JSON content-type unless you set one); a string/Uint8Array is sent as-is. Relative paths resolve against options.origin (default http://nifra.local). Body text is captured up to options.maxBodyChars (default 64 KB).

It's a runner, not a security sandbox: it calls the app you hand it in-process, with no isolation. Isolation, when you need it, comes from the host - the browser tab for the playground, your own process/CI for the agent runner. Don't feed it code you wouldn't already run.

For AI agents

Start with LLM.md - this package's contract card (the exports you call + its footguns), one cheap read instead of the whole corpus. For the wider framework: the repo's AGENTS.md is the copy-paste quick reference, and llms-full.txt is the full machine-readable corpus. Run nifra check as the done-gate, or nifra mcp to give the agent live project tools.