katt 0.0.4 → 0.0.6

package/README.md

@@ -1,20 +1,44 @@
 # Katt
+[![GitHub license](https://img.shields.io/badge/license-MIT-blue.svg)](https://github.com/raphaelpor/katt/blob/main/LICENSE) [![npm version](https://img.shields.io/npm/v/katt.svg?style=flat)](https://www.npmjs.com/package/katt)
 
 Katt is a lightweight testing framework for running AI Evals, inspired by [Jest](https://github.com/jestjs/jest).
 
 <img src="https://raw.githubusercontent.com/raphaelpor/katt/main/docs/logo.png" alt="Katt logo" width="250" />
 
+## Table of Contents
+
+- [Overview](#overview)
+- [API Documentation](#api-documentation)
+- [Hello World - Example](#hello-world---example)
+- [Main Features](#main-features)
+- [Usage](#usage)
+- [Installation](#installation)
+- [Basic Usage](#basic-usage)
+- [Using promptFile](#using-promptfile)
+- [Specifying AI Models](#specifying-ai-models)
+- [Development](#development)
+- [Setup](#setup)
+- [Available Scripts](#available-scripts)
+- [Verification Process](#verification-process)
+- [Project Structure](#project-structure)
+- [How It Works](#how-it-works)
+- [Requirements](#requirements)
+- [License](#license)
+- [Contributing](#contributing)
+
 ## Overview
 
 Katt is designed to evaluate and validate the behavior of AI agents like **Claude Code**, **GitHub Copilot**, **OpenAI Codex** and more. It provides a simple, intuitive API for writing tests that interact with AI models and assert their responses.
 
 ## API Documentation
 
-For a complete list of features and usage examples, see [docs/api-documentation.md](docs/api-documentation.md).
+For a complete list of features and usage examples, see [docs/api-documentation.md](https://github.com/raphaelpor/katt/blob/main/docs/api-documentation.md).
 
 ## Hello World - Example
 
 ```typescript
+import { expect, prompt } from "katt";
+
 const result = await prompt("If you read this just say 'hello world'");
 expect(result).toContain("hello world");
 ```
@@ -22,6 +46,8 @@ expect(result).toContain("hello world");
 It also supports the familiar `describe` and `it` syntax for organizing tests:
 
 ```typescript
+import { describe, expect, it, prompt } from "katt";
+
 describe("Greeting agent", () => {
   it("should say hello world", async () => {
     const result = await prompt("If you read this just say 'hello world'");
@@ -41,10 +67,18 @@ describe("Greeting agent", () => {
 
 ## Usage
 
+### Installation
+
+```bash
+npm install -g katt
+```
+
 ### Basic Usage
 
 1. Create a file with the `.eval.ts` or `.eval.js` extension and write your tests.
 ```typescript
+import { expect, prompt } from "katt";
+
 const result = await prompt("If you read this just say 'hello world'");
 expect(result).toContain("hello world");
 ```
@@ -52,7 +86,7 @@ expect(result).toContain("hello world");
 2. Run Katt from your project directory:
 
 ```bash
-npx katt
+katt
 ```
 
 ### Using promptFile
@@ -61,6 +95,8 @@ Load prompts from external files:
 
 ```javascript
 // test.eval.js
+import { describe, expect, it, promptFile } from "katt";
+
 describe("Working with files", () => {
   it("should load the file and respond", async () => {
     const result = await promptFile("./myPrompt.md");
@@ -74,6 +110,8 @@ describe("Working with files", () => {
 You can specify a custom model for your prompts:
 
 ```javascript
+import { describe, expect, it, prompt } from "katt";
+
 describe("Model selection", () => {
   it("should use a specific model", async () => {
     const promptString = "You are a helpful agent. Say hi and ask what you could help the user with.";
@@ -167,7 +205,7 @@ MIT
 
 ## Contributing
 
-We welcome contributions from the community! Please see our [CONTRIBUTING.md](CONTRIBUTING.md) guide for detailed information on how to contribute to Katt.
+We welcome contributions from the community! Please see our [CONTRIBUTING.md](https://github.com/raphaelpor/katt/blob/main/CONTRIBUTING.md) guide for detailed information on how to contribute to Katt.
 
 Quick start:
 1. Fork the repository
@@ -176,4 +214,4 @@ Quick start:
 4. Run the verification process
 5. Submit a pull request
 
-For detailed guidelines, development setup, coding standards, and more, check out our [contribution guide](CONTRIBUTING.md).
+For detailed guidelines, development setup, coding standards, and more, check out our [contribution guide](https://github.com/raphaelpor/katt/blob/main/CONTRIBUTING.md).

package/dist/index.js

@@ -1,183 +1,56 @@
-#!/usr/bin/env node
-import { AsyncLocalStorage as z } from "node:async_hooks";
-import { CopilotClient as it } from "@github/copilot-sdk";
-import { readFile as X, readdir as st } from "node:fs/promises";
-import { resolve as J, dirname as R, isAbsolute as rt, basename as ct, join as at } from "node:path";
-import { readFileSync as V, writeFileSync as W, mkdirSync as lt } from "node:fs";
-import { fileURLToPath as ut, pathToFileURL as dt } from "node:url";
-const Y = new z(), ft = {
-  describeStack: [],
-  itStack: [],
-  tokenUsageStack: [],
-  modelStack: []
-};
-let Z = 0, E = 0;
-const w = [], y = [];
-let U = 0;
-function u() {
-  return Y.getStore() ?? ft;
-}
-function Q(t) {
-  return {
-    describeStack: [...t.describeStack],
-    itStack: [...t.itStack],
-    tokenUsageStack: [...t.tokenUsageStack],
-    modelStack: [...t.modelStack]
-  };
-}
-function gt() {
-  return Z += 1, `d${Z}`;
-}
-function pt() {
-  return E += 1, `i${E}`;
-}
-function K(t, e) {
-  const n = e ?? Q(u());
-  return Y.run(n, t);
-}
-function _() {
-  return Q(u());
-}
-function ht(t) {
-  u().describeStack.push({ id: gt(), description: t });
-}
-function x() {
-  u().describeStack.pop();
-}
-function q() {
-  return u().describeStack.map((t) => t.description).join(" > ");
-}
-function It(t) {
-  u().itStack.push({ id: pt(), description: t }), u().tokenUsageStack.push(0), u().modelStack.push(void 0);
-}
-function F() {
-  u().itStack.pop(), u().tokenUsageStack.pop(), u().modelStack.pop();
-}
-function tt() {
-  return u().itStack.map((t) => t.description).join(" > ");
-}
-function mt(t) {
-  if (!Number.isFinite(t) || t <= 0)
-    return;
-  const e = u(), n = e.tokenUsageStack.length - 1;
-  n < 0 || (e.tokenUsageStack[n] += t);
-}
-function Ct() {
-  const t = u(), e = t.tokenUsageStack.length - 1;
-  return e < 0 ? 0 : t.tokenUsageStack[e] ?? 0;
-}
-function St(t) {
-  if (t.length === 0)
-    return;
-  const e = u(), n = e.modelStack.length - 1;
-  n < 0 || (e.modelStack[n] = t);
-}
-function At() {
-  const t = u(), e = t.modelStack.length - 1;
-  if (!(e < 0))
-    return t.modelStack[e];
-}
-function v(t) {
-  w.push(t);
-}
-function bt() {
-  U += 1;
-}
-function $t() {
-  return U;
-}
-function kt() {
-  U = 0;
-}
-function wt(t) {
-  y.push(t);
-}
-function yt() {
-  return [...y];
-}
-function O() {
-  return y.length;
-}
-function vt() {
-  y.length = 0;
-}
-async function Lt() {
-  const t = [];
-  for (; w.length > 0; ) {
-    const e = w.splice(0, w.length), n = await Promise.allSettled(e);
-    t.push(...n);
-  }
-  return t;
+import { r as $, c as x, p as E, a as d, l, b as p, g as w, d as N, e as P, f as h, s as A, h as S, i as j, j as _, k as O, m as L, n as I, o as U, q as R } from "./runCli-B3oIBxOl.js";
+import { t as xt } from "./runCli-B3oIBxOl.js";
+import { CopilotClient as z } from "@github/copilot-sdk";
+import { readFile as F } from "node:fs/promises";
+import { resolve as k, dirname as g, isAbsolute as B, basename as W, join as K } from "node:path";
+import { readFileSync as H, writeFileSync as C, mkdirSync as Y } from "node:fs";
+function wt(t, e) {
+  $(() => {
+    N(), E(t);
+    const n = w(), o = Date.now(), r = () => w() === n, s = () => Date.now() - o;
+    try {
+      const i = e();
+      if (i && typeof i.then == "function") {
+        d(
+          i.then(() => {
+            l(!0, s());
+          }).catch((a) => {
+            throw l(!1, s()), a;
+          }).finally(() => {
+            p();
+          })
+        );
+        return;
+      }
+    } catch (i) {
+      throw l(!1, s()), p(), i;
+    }
+    l(r(), s()), p();
+  }, x());
 }
-function jt(t, e) {
-  K(() => {
-    ht(t);
+function Ct(t, e) {
+  $(() => {
+    P(t);
     try {
       const n = e();
       if (n && typeof n.then == "function") {
-        v(
+        d(
           n.finally(() => {
-            x();
+            h();
           })
         );
         return;
       }
     } catch (n) {
-      throw x(), n;
+      throw h(), n;
     }
-    x();
-  }, _());
-}
-const Tt = "\x1B[1;36m", xt = "\x1B[33m", Ft = "\x1B[38;5;208m", Mt = "\x1B[1;38;5;208m", L = "\x1B[0m";
-function f(t) {
-  return `${Tt}${t}${L}`;
-}
-function $(t) {
-  return `${xt}${t}${L}`;
-}
-function D(t) {
-  return `${Ft}${t}${L}`;
-}
-function Nt(t) {
-  return `${Mt}${t}${L}`;
-}
-let B = "";
-function Bt() {
-  B = "";
+    h();
+  }, x());
 }
-function Jt({
-  suitePath: t,
-  casePath: e,
-  didPass: n,
-  durationMs: o,
-  model: s,
-  tokenUsage: i
-}) {
-  const r = t.length > 0 ? t : "(root)", c = e.length > 0 ? e : "(root)";
-  B !== r && (console.log(`Suite "${f(r)}"`), B = r);
-  const l = n ? "✅ Passed in" : "❌ Failed in", I = [
-    `Test "${f(c)}"`,
-    `- ${l} ${f(`${o}ms`)}`
-  ];
-  s && I.push(`- Model ${f(s)}`), (i ?? 0) > 0 && I.push(`- Tokens used ${f(String(i))}`), I.push("---"), console.log(I.join(`
-`));
-}
-function C(t, e, n = "(root)") {
-  const o = tt();
-  Jt({
-    suitePath: q(),
-    casePath: o.length > 0 ? o : n,
-    didPass: t,
-    durationMs: e,
-    model: At(),
-    tokenUsage: Ct()
-  });
-}
-const G = new z();
-function Rt(t, e) {
+function q(t, e) {
   return typeof t == "object" && t !== null && "code" in t && t.code === e;
 }
-function Ut(t) {
+function J(t) {
   try {
     const e = JSON.parse(t);
     return typeof e == "object" && e !== null ? e : void 0;
@@ -186,19 +59,19 @@ function Ut(t) {
     return;
   }
 }
-async function Gt() {
-  const t = J(process.cwd(), "katt.json");
+async function G() {
+  const t = k(process.cwd(), "katt.json");
   try {
-    const e = await X(t, "utf8");
-    return Ut(e);
+    const e = await F(t, "utf8");
+    return J(e);
   } catch (e) {
-    if (Rt(e, "ENOENT"))
+    if (q(e, "ENOENT"))
       return;
     console.warn(`Failed to read katt.json: ${String(e)}`);
     return;
   }
 }
-function Wt(t) {
+function Q(t) {
   const e = t?.copilot;
   if (typeof e != "object" || e === null || Array.isArray(e))
     return;
@@ -207,112 +80,134 @@ function Wt(t) {
   }, o = n.model;
   return (typeof o != "string" || o.length === 0) && delete n.model, Object.keys(n).length > 0 ? n : void 0;
 }
-function Zt(t) {
+function V(t) {
   if (!(typeof t != "number" || !Number.isFinite(t)) && !(t <= 0))
     return Math.floor(t);
 }
-function Et(t) {
+function X(t) {
   const e = t?.prompt;
   if (!(typeof e != "object" || e === null || Array.isArray(e)))
-    return Zt(e.timeoutMs);
+    return V(e.timeoutMs);
 }
-async function Ot() {
-  const t = await Gt();
+async function Z() {
+  const t = await G();
   return {
-    copilot: Wt(t),
-    promptTimeoutMs: Et(t)
+    copilot: Q(t),
+    promptTimeoutMs: X(t)
   };
 }
-const Dt = 6e5;
-function et(t) {
+const tt = 6e5;
+function M(t) {
   return typeof t == "string" && t.length > 0 ? t : void 0;
 }
-function M(t) {
+function m(t) {
   if (!t)
     return;
   const e = { ...t };
   if (e.model !== void 0) {
-    const n = et(e.model);
+    const n = M(e.model);
     n ? e.model = n : delete e.model;
   }
   return Object.keys(e).length > 0 ? e : void 0;
 }
-function H(t) {
+function b(t) {
   if (!(typeof t != "number" || !Number.isFinite(t)) && !(t <= 0))
     return Math.floor(t);
 }
-function k(t) {
+function f(t) {
   return !Number.isFinite(t) || (t ?? 0) <= 0 ? 0 : Math.floor(t ?? 0);
 }
-function Ht(t) {
-  return k(t.inputTokens) + k(t.outputTokens) + k(t.cacheReadTokens) + k(t.cacheWriteTokens);
+function et(t) {
+  return f(t.inputTokens) + f(t.outputTokens) + f(t.cacheReadTokens) + f(t.cacheWriteTokens);
 }
-async function j(t, e = {}) {
-  const { timeoutMs: n, ...o } = e, s = await Ot(), i = M(s.copilot), r = M(
+function nt(t, e) {
+  const { timeoutMs: n, ...o } = t, r = m(e.copilot), s = m(
     o
-  ), c = M({
-    ...i ?? {},
-    ...r ?? {}
-  }), l = H(s.promptTimeoutMs), T = H(n) ?? l ?? Dt, a = et(c?.model), d = new it({ useLoggedInUser: !0 });
-  let p, b, S = 0;
-  try {
-    await d.start(), p = await d.createSession(c), b = p.on("assistant.usage", (m) => {
-      S += Ht(m.data);
-    });
-    const h = await p.sendAndWait({ prompt: t }, T);
-    if (!h?.data?.content)
-      throw new Error("Copilot did not return a response.");
-    return a && St(a), h.data.content;
-  } finally {
-    const h = [];
-    if (b?.(), S > 0 && mt(S), p)
-      try {
-        await p.destroy();
-      } catch (m) {
-        h.push(m);
-      }
+  ), i = m({
+    ...r ?? {},
+    ...s ?? {}
+  }), a = b(e.promptTimeoutMs), D = b(n) ?? a ?? tt;
+  return {
+    sessionOptions: i,
+    model: M(i?.model),
+    timeoutMs: D
+  };
+}
+async function ot(t, e) {
+  const n = [];
+  if (e.unsubscribeUsage?.(), e.usedTokens > 0 && j(e.usedTokens), e.session)
     try {
-      const m = await d.stop();
-      h.push(...m);
-    } catch (m) {
-      h.push(m);
+      await e.session.destroy();
+    } catch (o) {
+      n.push(o);
     }
-    h.length > 0 && console.error(
-      `Copilot cleanup encountered ${h.length} error(s).`
+  try {
+    const o = await t.stop();
+    n.push(...o);
+  } catch (o) {
+    n.push(o);
+  }
+  n.length > 0 && console.error(
+    `Copilot cleanup encountered ${n.length} error(s).`
+  );
+}
+async function y(t, e = {}) {
+  const n = await Z(), o = nt(e, n), r = new z({ useLoggedInUser: !0 }), s = {
+    session: void 0,
+    unsubscribeUsage: void 0,
+    usedTokens: 0
+  };
+  try {
+    await r.start(), s.session = await r.createSession(
+      o.sessionOptions
+    ), s.unsubscribeUsage = s.session.on(
+      "assistant.usage",
+      (a) => {
+        s.usedTokens += et(a.data);
+      }
+    );
+    const i = await s.session.sendAndWait(
+      { prompt: t },
+      o.timeoutMs
     );
+    if (!i?.data?.content)
+      throw new Error("Copilot did not return a response.");
+    return o.model && A(o.model), i.data.content;
+  } finally {
+    await ot(r, s);
   }
 }
-async function Pt(t, e = {}) {
-  const n = G.getStore(), o = n?.evalFile ? R(n.evalFile) : process.cwd(), s = rt(t) ? t : J(o, t), i = await X(s, "utf8");
-  return j(i, e);
+async function bt(t, e = {}) {
+  const n = S.getStore(), o = n?.evalFile ? g(n.evalFile) : process.cwd(), r = B(t) ? t : k(o, t), s = await F(r, "utf8");
+  return y(s, e);
 }
-function g(t) {
-  wt({
-    describePath: q(),
-    itPath: tt(),
+function c(t) {
+  _({
+    describePath: L(),
+    itPath: O(),
     message: t
   });
 }
-async function zt(t, e) {
-  const n = Date.now(), o = `expected '${t}' to satisfy '${e}'`, s = j(`Evaluate if the expectation is fulfiled in by the input.
+async function st(t, e) {
+  const n = Date.now(), o = `expected '${t}' to satisfy '${e}'`, r = y(`Evaluate if the expectation is fulfiled in by the input.
         Expectation: "${e}".
         Input:
         ---
         ${t}
         ---
         Important: Answer with "Yes" or "No" only, without any additional text.
-        `).then((i) => {
-    i.includes("Yes") ? C(!0, Date.now() - n, "promptCheck") : i.includes("No") ? g(o) : g(`failed to evaluate expectation '${e}'`);
+        `).then((s) => {
+    s.includes("Yes") ? l(!0, Date.now() - n, "promptCheck") : s.includes("No") ? c(o) : c(`failed to evaluate expectation '${e}'`);
   });
-  return v(s), s;
+  return d(r), r;
 }
-const Xt = 3;
-function Vt(t) {
+const rt = 3;
+function it(t) {
   const e = t.match(/\b([1-5])\b/);
   return !e || !e[1] ? null : Number(e[1]);
 }
-async function Yt(t, e, n) {
-  const o = Date.now(), s = n?.threshold ?? Xt, i = j(
+async function at(t, e, n) {
+  const o = Date.now(), r = n?.threshold ?? rt, s = y(
     `Classify the input by how "${e}" it is on a scale of 1 to 5.
 1 means "not ${e}" and 5 means "very ${e}".
 Return only a single number: 1, 2, 3, 4, or 5.
@@ -322,94 +217,99 @@ Input:
 ${t}
 ---`,
     n?.model ? { model: n.model } : void 0
-  ).then((r) => {
-    const c = Vt(r);
-    if (c === null) {
-      g(
-        `failed to classify as '${e}'. Evaluator returned '${r}'`
+  ).then((i) => {
+    const a = it(i);
+    if (a === null) {
+      c(
+        `failed to classify as '${e}'. Evaluator returned '${i}'`
       );
       return;
     }
-    const l = `expected response to be classified as '${e}' with score >= ${s}, got ${c}`;
-    if (c < s) {
-      g(l);
+    const u = `expected response to be classified as '${e}' with score >= ${r}, got ${a}`;
+    if (a < r) {
+      c(u);
       return;
     }
-    C(
+    l(
       !0,
       Date.now() - o,
       "toBeClassifiedAs"
     );
   });
-  return v(i), i;
+  return d(s), s;
 }
-function Qt(t, e) {
+function ct(t, e) {
   const n = `expected '${t}' to include '${e}'`;
-  t.includes(e) || g(n);
-}
-let nt = !1;
-function Kt(t) {
-  nt = t;
-}
-function _t() {
-  return nt;
+  t.includes(e) || c(n);
 }
-function qt(t) {
-  const n = ct(t).replace(/\.eval\.[^./\\]+$/, "");
-  return at(
-    R(t),
+function v(t) {
+  const e = t.trim().replace(/[<>:"/\\|?*\x00-\x1f]/g, "_").replace(/\s+/g, "_");
+  return e.length > 0 ? e : "unnamed";
+}
+function ut() {
+  const t = U().map(
+    (o) => v(o.description)
+  ), e = R().map(
+    (o) => v(o.description)
+  ), n = [...t, ...e];
+  return n.length === 0 ? "root" : n.join("__");
+}
+function lt(t) {
+  const n = W(t).replace(/\.eval\.[^./\\]+$/, ""), o = ut();
+  return K(
+    g(t),
     "__snapshots__",
-    `${n}.snap.md`
+    `${n}__${o}.snap.md`
   );
 }
-function P(t) {
+function T(t) {
   return t.split(/\r?\n/);
 }
-function te(t, e) {
+function ft(t, e) {
   if (t === e)
     return "  (no diff)";
-  const n = P(t), o = P(e), s = Math.max(n.length, o.length), i = [];
-  for (let r = 0; r < s; r += 1) {
-    const c = n[r], l = o[r];
-    if (c !== l) {
-      if (c === void 0 && l !== void 0) {
-        i.push(`+ ${l}`);
+  const n = T(t), o = T(e), r = Math.max(n.length, o.length), s = [];
+  for (let i = 0; i < r; i += 1) {
+    const a = n[i], u = o[i];
+    if (a !== u) {
+      if (a === void 0 && u !== void 0) {
+        s.push(`+ ${u}`);
         continue;
       }
-      if (c !== void 0 && l === void 0) {
-        i.push(`- ${c}`);
+      if (a !== void 0 && u === void 0) {
+        s.push(`- ${a}`);
         continue;
       }
-      i.push(`- ${c ?? ""}`), i.push(`+ ${l ?? ""}`);
+      s.push(`- ${a ?? ""}`), s.push(`+ ${u ?? ""}`);
     }
   }
-  return i.join(`
+  return s.join(`
 `);
 }
-function ee(t) {
-  const e = G.getStore()?.evalFile;
+function dt(t) {
+  const e = S.getStore()?.evalFile;
   if (!e) {
-    g(
+    c(
       "toMatchSnapshot can only be used while running an eval file."
     );
     return;
   }
-  const n = qt(e);
+  const n = lt(e);
   try {
-    const o = V(n, "utf8");
+    const o = H(n, "utf8");
     if (o === t)
       return;
-    if (_t()) {
-      W(n, t, "utf8");
+    if (I()) {
+      C(n, t, "utf8");
       return;
     }
-    const s = te(o, t);
-    g(
+    const r = ft(o, t);
+    c(
       [
         `Snapshot mismatch at ${n}`,
         "",
         "Diff:",
-        s,
+        r,
         "",
         "Run katt with --update-snapshots (or -u) to accept this change."
       ].join(`
@@ -417,173 +317,41 @@ function ee(t) {
     );
   } catch (o) {
     if (o.code !== "ENOENT") {
-      g(
+      c(
         `Failed to read snapshot at ${n}: ${String(o)}`
       );
       return;
     }
     try {
-      lt(R(n), { recursive: !0 }), W(n, t, "utf8");
-    } catch (i) {
-      g(
-        `Failed to write snapshot at ${n}: ${String(i)}`
+      Y(g(n), { recursive: !0 }), C(n, t, "utf8");
+    } catch (s) {
+      c(
+        `Failed to write snapshot at ${n}: ${String(s)}`
       );
     }
   }
 }
-function ne(t) {
+function vt(t) {
   return {
     toContain: (e) => {
-      Qt(t, e);
+      ct(t, e);
     },
     toMatchSnapshot: () => {
-      ee(t);
+      dt(t);
     },
     promptCheck: async (e) => {
-      await zt(t, e);
+      await st(t, e);
     },
     toBeClassifiedAs: async (e, n) => {
-      await Yt(t, e, n);
+      await at(t, e, n);
     }
   };
 }
-function oe(t, e) {
-  K(() => {
-    bt(), It(t);
-    const n = O(), o = Date.now(), s = () => O() === n, i = () => Date.now() - o;
-    try {
-      const r = e();
-      if (r && typeof r.then == "function") {
-        v(
-          r.then(() => {
-            C(!0, i());
-          }).catch((c) => {
-            throw C(!1, i()), c;
-          }).finally(() => {
-            F();
-          })
-        );
-        return;
-      }
-    } catch (r) {
-      throw C(!1, i()), F(), r;
-    }
-    C(s(), i()), F();
-  }, _());
-}
-const ie = /\.eval\.(js|ts)$/, se = /* @__PURE__ */ new Set([".git", "node_modules"]);
-async function ot(t) {
-  const e = await st(t, { withFileTypes: !0 }), n = [];
-  return await Promise.all(
-    e.map(async (o) => {
-      const s = J(t, o.name);
-      if (o.isDirectory()) {
-        if (se.has(o.name))
-          return;
-        n.push(...await ot(s));
-        return;
-      }
-      o.isFile() && ie.test(o.name) && n.push(s);
-    })
-  ), n;
-}
-const N = new URL("data:application/json;base64,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", import.meta.url);
-let A;
-function re() {
-  if (A !== void 0)
-    return A;
-  try {
-    const t = N.protocol === "data:" ? ce(N) : V(ut(N), "utf8"), e = JSON.parse(t);
-    A = typeof e.version == "string" ? e.version : "unknown";
-  } catch {
-    A = "unknown";
-  }
-  return A;
-}
-function ce(t) {
-  const e = t.pathname.indexOf(",");
-  if (e < 0)
-    throw new Error("Invalid data URL.");
-  const n = t.pathname.slice(0, e), o = t.pathname.slice(e + 1);
-  return n.includes(";base64") ? Buffer.from(o, "base64").toString("utf8") : decodeURIComponent(o);
-}
-function ae() {
-  const t = " ██╗  ██╗ █████╗ ████████╗████████╗", e = " ██║ ██╔╝██╔══██╗╚══██╔══╝╚══██╔══╝", n = " █████╔╝ ███████║   ██║      ██║", o = " ██╔═██╗ ██╔══██║   ██║      ██║", s = " ██║  ██╗██║  ██║   ██║      ██║", i = " ╚═╝  ╚═╝╚═╝  ╚═╝   ╚═╝      ╚═╝", r = `v${re()}`, c = Math.max(
-    0,
-    Math.floor((t.length - r.length) / 2)
-  ), l = `${" ".repeat(c)}${r}`;
-  console.log(`
-${$(t)}
-${$(e)}
-${$(n)}
-${D(o)}
-${D(s)}
-${Nt(i)}
-${$(l)}
-`);
-}
-function le(t) {
-  const e = String(t.getHours()).padStart(2, "0"), n = String(t.getMinutes()).padStart(2, "0"), o = String(t.getSeconds()).padStart(2, "0");
-  return `${e}:${n}:${o}`;
-}
-async function ue() {
-  const t = process.argv.slice(2), e = t.includes("--update-snapshots") || t.includes("-u");
-  Kt(e), ae();
-  const n = /* @__PURE__ */ new Date();
-  Bt(), vt(), kt();
-  const o = await ot(process.cwd());
-  if (o.length === 0)
-    return console.log("No .eval.js or .eval.ts files found."), 1;
-  const i = (await Promise.allSettled(
-    o.map(
-      (a) => G.run(
-        { evalFile: a },
-        () => import(dt(a).href)
-      )
-    )
-  )).map((a, d) => ({ result: a, file: o[d] })).filter(({ result: a }) => a.status === "rejected");
-  if (i.length > 0) {
-    for (const a of i) {
-      const d = a.result.status === "rejected" ? a.result.reason : void 0;
-      console.error(`Error executing ${a.file}: ${String(d)}`);
-    }
-    return 1;
-  }
-  const c = (await Lt()).filter(
-    (a) => a.status === "rejected"
-  );
-  if (c.length > 0) {
-    for (const a of c)
-      a.status === "rejected" && console.error(`Error executing async test: ${String(a.reason)}`);
-    return 1;
-  }
-  const l = yt();
-  if (l.length > 0) {
-    console.error("❌ Failed tests:");
-    for (const [a, d] of l.entries()) {
-      const p = [d.describePath, d.itPath].filter((S) => S.length > 0).join(" > "), b = p.length > 0 ? `${p}: ` : "";
-      console.error(`${a + 1}. ${b}${d.message}`);
-    }
-    return 1;
-  }
-  const I = $t(), T = Date.now() - n.getTime();
-  return console.log(
-    [
-      "---",
-      `${f("Files")}  ${o.length} passed`,
-      `${f("Evals")}  ${I} passed`,
-      `${f("Start at")}  ${le(n)}`,
-      `${f("Duration")}  ${T}ms`
-    ].join(`
-`)
-  ), 0;
-}
-Object.assign(globalThis, { describe: jt, it: oe, expect: ne, prompt: j, promptFile: Pt });
-ue().then((t) => {
-  process.exit(t);
-}).catch((t) => {
-  console.error(`Unexpected error: ${String(t)}`), process.exit(1);
-});
 export {
-  ue as runCli
+  Ct as describe,
+  vt as expect,
+  wt as it,
+  y as prompt,
+  bt as promptFile,
+  xt as runCli
 };

package/dist/katt.js

@@ -0,0 +1,7 @@
+#!/usr/bin/env node
+import { t as r } from "./runCli-B3oIBxOl.js";
+r().then((e) => {
+  process.exit(e);
+}).catch((e) => {
+  console.error(`Unexpected error: ${String(e)}`), process.exit(1);
+});

package/dist/runCli-B3oIBxOl.js

@@ -0,0 +1,317 @@
+import { fileURLToPath as Z, pathToFileURL as y } from "node:url";
+import { readdir as N } from "node:fs/promises";
+import { resolve as M } from "node:path";
+import { AsyncLocalStorage as v } from "node:async_hooks";
+import { readFileSync as X } from "node:fs";
+const H = /\.eval\.(js|ts)$/, R = /* @__PURE__ */ new Set([".git", "node_modules"]);
+async function w(t) {
+  const e = await N(t, { withFileTypes: !0 }), n = [];
+  return await Promise.all(
+    e.map(async (i) => {
+      const s = M(t, i.name);
+      if (i.isDirectory()) {
+        if (R.has(i.name))
+          return;
+        n.push(...await w(s));
+        return;
+      }
+      i.isFile() && H.test(i.name) && n.push(s);
+    })
+  ), n;
+}
+const j = new v(), W = {
+  describeStack: [],
+  itStack: [],
+  tokenUsageStack: [],
+  modelStack: []
+};
+let b = 0, L = 0;
+const p = [], f = [];
+let k = 0;
+function o() {
+  return j.getStore() ?? W;
+}
+function J(t) {
+  return {
+    describeStack: [...t.describeStack],
+    itStack: [...t.itStack],
+    tokenUsageStack: [...t.tokenUsageStack],
+    modelStack: [...t.modelStack]
+  };
+}
+function U() {
+  return b += 1, `d${b}`;
+}
+function G() {
+  return L += 1, `i${L}`;
+}
+function B(t) {
+  return t.tokenUsageStack.length - 1;
+}
+function F(t) {
+  return t.modelStack.length - 1;
+}
+function bt(t, e) {
+  const n = e ?? J(o());
+  return j.run(n, t);
+}
+function Lt() {
+  return J(o());
+}
+function $t(t) {
+  o().describeStack.push({ id: U(), description: t });
+}
+function vt() {
+  o().describeStack.pop();
+}
+function O() {
+  return o().describeStack.map((t) => t.description).join(" > ");
+}
+function wt() {
+  return [...o().describeStack];
+}
+function jt(t) {
+  const e = o();
+  e.itStack.push({ id: G(), description: t }), e.tokenUsageStack.push(0), e.modelStack.push(void 0);
+}
+function Jt() {
+  const t = o();
+  t.itStack.pop(), t.tokenUsageStack.pop(), t.modelStack.pop();
+}
+function Y() {
+  return o().itStack.map((t) => t.description).join(" > ");
+}
+function Bt() {
+  return [...o().itStack];
+}
+function Ft(t) {
+  if (!Number.isFinite(t) || t <= 0)
+    return;
+  const e = o(), n = B(e);
+  n < 0 || (e.tokenUsageStack[n] += t);
+}
+function Q() {
+  const t = o(), e = B(t);
+  return e < 0 ? 0 : t.tokenUsageStack[e] ?? 0;
+}
+function xt(t) {
+  if (t.length === 0)
+    return;
+  const e = o(), n = F(e);
+  n < 0 || (e.modelStack[n] = t);
+}
+function V() {
+  const t = o(), e = F(t);
+  if (!(e < 0))
+    return t.modelStack[e];
+}
+function Tt(t) {
+  p.push(t);
+}
+function Zt() {
+  k += 1;
+}
+function K() {
+  return k;
+}
+function P() {
+  k = 0;
+}
+function yt(t) {
+  f.push(t);
+}
+function E() {
+  return [...f];
+}
+function Nt() {
+  return f.length;
+}
+function z() {
+  f.length = 0;
+}
+async function D() {
+  const t = [];
+  for (; p.length > 0; ) {
+    const e = p.splice(0, p.length), n = await Promise.allSettled(e);
+    t.push(...n);
+  }
+  return t;
+}
+const _ = "\x1B[1;36m", q = "\x1B[33m", tt = "\x1B[38;5;208m", et = "\x1B[1;38;5;208m", m = "\x1B[0m";
+function r(t) {
+  return `${_}${t}${m}`;
+}
+function C(t) {
+  return `${q}${t}${m}`;
+}
+function $(t) {
+  return `${tt}${t}${m}`;
+}
+function nt(t) {
+  return `${et}${t}${m}`;
+}
+let A = "";
+function it() {
+  A = "";
+}
+function ot({
+  suitePath: t,
+  casePath: e,
+  didPass: n,
+  durationMs: i,
+  model: s,
+  tokenUsage: c
+}) {
+  const l = t.length > 0 ? t : "(root)", u = e.length > 0 ? e : "(root)";
+  A !== l && (console.log(`Suite "${r(l)}"`), A = l);
+  const g = n ? "✅ Passed in" : "❌ Failed in", I = [
+    `Test "${r(u)}"`,
+    `- ${g} ${r(`${i}ms`)}`
+  ];
+  s && I.push(`- Model ${r(s)}`), (c ?? 0) > 0 && I.push(`- Tokens used ${r(String(c))}`), I.push("---"), console.log(I.join(`
+`));
+}
+function Mt(t, e, n = "(root)") {
+  const i = Y();
+  ot({
+    suitePath: O(),
+    casePath: i.length > 0 ? i : n,
+    didPass: t,
+    durationMs: e,
+    model: V(),
+    tokenUsage: Q()
+  });
+}
+const st = new v(), h = new URL("data:application/json;base64,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", import.meta.url);
+let d;
+function at() {
+  if (d !== void 0)
+    return d;
+  try {
+    const t = h.protocol === "data:" ? rt(h) : X(Z(h), "utf8"), e = JSON.parse(t);
+    d = typeof e.version == "string" ? e.version : "unknown";
+  } catch {
+    d = "unknown";
+  }
+  return d;
+}
+function rt(t) {
+  const e = t.pathname.indexOf(",");
+  if (e < 0)
+    throw new Error("Invalid data URL.");
+  const n = t.pathname.slice(0, e), i = t.pathname.slice(e + 1);
+  return n.includes(";base64") ? Buffer.from(i, "base64").toString("utf8") : decodeURIComponent(i);
+}
+function ct() {
+  const t = " ██╗  ██╗ █████╗ ████████╗████████╗", e = " ██║ ██╔╝██╔══██╗╚══██╔══╝╚══██╔══╝", n = " █████╔╝ ███████║   ██║      ██║", i = " ██╔═██╗ ██╔══██║   ██║      ██║", s = " ██║  ██╗██║  ██║   ██║      ██║", c = " ╚═╝  ╚═╝╚═╝  ╚═╝   ╚═╝      ╚═╝", l = `v${at()}`, u = Math.max(
+    0,
+    Math.floor((t.length - l.length) / 2)
+  ), g = `${" ".repeat(u)}${l}`;
+  console.log(`
+${C(t)}
+${C(e)}
+${C(n)}
+${$(i)}
+${$(s)}
+${nt(c)}
+${C(g)}
+`);
+}
+let x = !1;
+function lt(t) {
+  x = t;
+}
+function Xt() {
+  return x;
+}
+function ut(t) {
+  const e = String(t.getHours()).padStart(2, "0"), n = String(t.getMinutes()).padStart(2, "0"), i = String(t.getSeconds()).padStart(2, "0");
+  return `${e}:${n}:${i}`;
+}
+function gt(t) {
+  return t.includes("--update-snapshots") || t.includes("-u");
+}
+function It(t) {
+  return [t.describePath, t.itPath].filter((e) => e.length > 0).join(" > ");
+}
+function dt(t) {
+  for (const e of t)
+    console.error(`Error executing ${e.file}: ${String(e.reason)}`);
+}
+function Ct(t) {
+  for (const e of t)
+    console.error(`Error executing async test: ${String(e.reason)}`);
+}
+function pt(t) {
+  console.error("❌ Failed tests:");
+  for (const [e, n] of t.entries()) {
+    const i = It(n), s = i.length > 0 ? `${i}: ` : "";
+    console.error(`${e + 1}. ${s}${n.message}`);
+  }
+}
+function ft(t, e, n, i) {
+  return [
+    "---",
+    `${r("Files")}  ${t} passed`,
+    `${r("Evals")}  ${e} passed`,
+    `${r("Start at")}  ${ut(n)}`,
+    `${r("Duration")}  ${i}ms`
+  ].join(`
+`);
+}
+async function Ht() {
+  const t = process.argv.slice(2), e = gt(t);
+  lt(e), ct();
+  const n = /* @__PURE__ */ new Date();
+  it(), z(), P();
+  const i = await w(process.cwd());
+  if (i.length === 0)
+    return console.log("No .eval.js or .eval.ts files found."), 1;
+  const c = (await Promise.allSettled(
+    i.map(
+      (a) => st.run(
+        { evalFile: a },
+        () => import(y(a).href)
+      )
+    )
+  )).map((a, S) => ({ result: a, file: i[S] })).filter(({ result: a }) => a.status === "rejected").map(({ result: a, file: S }) => ({
+    file: S,
+    reason: a.status === "rejected" ? a.reason : void 0
+  }));
+  if (c.length > 0)
+    return dt(c), 1;
+  const u = (await D()).filter(
+    (a) => a.status === "rejected"
+  );
+  if (u.length > 0)
+    return Ct(u), 1;
+  const g = E();
+  if (g.length > 0)
+    return pt(g), 1;
+  const I = K(), T = Date.now() - n.getTime();
+  return console.log(
+    ft(i.length, I, n, T)
+  ), 0;
+}
+export {
+  Tt as a,
+  Jt as b,
+  Lt as c,
+  Zt as d,
+  $t as e,
+  vt as f,
+  Nt as g,
+  st as h,
+  Ft as i,
+  yt as j,
+  Y as k,
+  Mt as l,
+  O as m,
+  Xt as n,
+  wt as o,
+  jt as p,
+  Bt as q,
+  bt as r,
+  xt as s,
+  Ht as t
+};

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "katt",
-  "version": "0.0.4",
+  "version": "0.0.6",
   "description": "CLI tool that tests the output of agentic AI tools",
   "keywords": [
     "cli",
@@ -13,8 +13,14 @@
   "license": "MIT",
   "type": "module",
   "main": "dist/index.js",
+  "exports": {
+    ".": {
+      "types": "./dist/index.d.ts",
+      "import": "./dist/index.js"
+    }
+  },
   "bin": {
-    "katt": "dist/index.js"
+    "katt": "dist/katt.js"
   },
   "scripts": {
     "build": "vite build",
@@ -23,7 +29,7 @@
     "format": "biome format --write ./src",
     "test": "vitest",
     "typecheck": "tsc -p tsconfig.json --noEmit",
-    "test:build": "node ./dist/index.js"
+    "test:build": "node ./dist/katt.js"
   },
   "types": "dist/index.d.ts",
   "devDependencies": {