@rudderjs/ai 1.4.0 → 1.6.0

package/dist/eval/fixtures.d.ts

@@ -0,0 +1,65 @@
+/**
+ * Fixture I/O for `pnpm rudder ai:eval --record` / `--replay` (#A5
+ * Phase 4). Each case writes one JSON file under
+ * `evals/__fixtures__/<suite>/<case>.json` carrying the assistant
+ * turns from a real provider run, normalized into the
+ * {@link AiFakeStep} shape so `--replay` can re-feed them via
+ * `AiFake.respondWithSequence` for zero-API regression tests.
+ *
+ * The fixture format is versioned. Bumping `version` forces a
+ * re-record on stale fixtures rather than silently mis-replaying.
+ */
+import type { AgentResponse } from '../types.js';
+import type { AiFakeStep } from '../fake.js';
+/** Fixture format. Bump `version` when the shape changes incompatibly. */
+export interface EvalFixture {
+    version: 1;
+    suite: string;
+    case: string;
+    input: string;
+    recordedAt: string;
+    steps: AiFakeStep[];
+}
+/**
+ * Convert an `AgentResponse` into the assistant-turn `AiFakeStep[]`
+ * sequence that `AiFake.respondWithSequence` expects.
+ *
+ * - Drops user/tool turns — those are framework-generated during a
+ *   replayed run, not provider output.
+ * - Multi-modal assistant content collapses to its concatenated text
+ *   parts (the fake's transport is text-only; image/document parts
+ *   wouldn't replay meaningfully).
+ * - `toolCalls` carry through verbatim so multi-step tool loops
+ *   replay deterministically.
+ */
+export declare function stepsFromResponse(response: AgentResponse): AiFakeStep[];
+/**
+ * Default fixtures directory: `<cwd>/evals/__fixtures__`. Override
+ * via the CLI handler's options for tests / non-standard layouts.
+ */
+export declare function defaultFixturesDir(cwd: string): string;
+/**
+ * Filesystem-safe slug for `<suite>/<case>` segments. Letters,
+ * digits, dot, dash, underscore pass through; everything else
+ * collapses to `-`. Multiple consecutive `-` collapse to one.
+ *
+ * Pure function; tested directly so suite/case rename diffs stay
+ * predictable across editors.
+ */
+export declare function slugify(s: string): string;
+export declare function fixturePath(dir: string, suite: string, caseName: string): string;
+/**
+ * Read a fixture file. Returns `null` when the fixture is missing
+ * (replay falls back to running normally with a clear stderr line).
+ *
+ * Throws on parse / version errors — corruption is not a passing
+ * case and silently ignoring it would mask real regressions.
+ */
+export declare function readFixture(dir: string, suite: string, caseName: string): Promise<EvalFixture | null>;
+/**
+ * Write a fixture, creating intermediate directories as needed.
+ * Pretty-printed (2-space) so PR diffs remain readable when the
+ * model output evolves.
+ */
+export declare function writeFixture(dir: string, suite: string, caseName: string, payload: Omit<EvalFixture, 'version' | 'suite' | 'case' | 'recordedAt'>): Promise<string>;
+//# sourceMappingURL=fixtures.d.ts.map

package/dist/eval/fixtures.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"fixtures.d.ts","sourceRoot":"","sources":["../../src/eval/fixtures.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAIH,OAAO,KAAK,EAAE,aAAa,EAAe,MAAM,aAAa,CAAA;AAC7D,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,YAAY,CAAA;AAE5C,0EAA0E;AAC1E,MAAM,WAAW,WAAW;IAC1B,OAAO,EAAK,CAAC,CAAA;IACb,KAAK,EAAO,MAAM,CAAA;IAClB,IAAI,EAAQ,MAAM,CAAA;IAClB,KAAK,EAAO,MAAM,CAAA;IAClB,UAAU,EAAE,MAAM,CAAA;IAClB,KAAK,EAAO,UAAU,EAAE,CAAA;CACzB;AAED;;;;;;;;;;;GAWG;AACH,wBAAgB,iBAAiB,CAAC,QAAQ,EAAE,aAAa,GAAG,UAAU,EAAE,CAWvE;AASD;;;GAGG;AACH,wBAAgB,kBAAkB,CAAC,GAAG,EAAE,MAAM,GAAG,MAAM,CAEtD;AAED;;;;;;;GAOG;AACH,wBAAgB,OAAO,CAAC,CAAC,EAAE,MAAM,GAAG,MAAM,CAEzC;AAED,wBAAgB,WAAW,CAAC,GAAG,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,GAAG,MAAM,CAEhF;AAID;;;;;;GAMG;AACH,wBAAsB,WAAW,CAC/B,GAAG,EAAO,MAAM,EAChB,KAAK,EAAK,MAAM,EAChB,QAAQ,EAAE,MAAM,GACf,OAAO,CAAC,WAAW,GAAG,IAAI,CAAC,CAiB7B;AAED;;;;GAIG;AACH,wBAAsB,YAAY,CAChC,GAAG,EAAO,MAAM,EAChB,KAAK,EAAK,MAAM,EAChB,QAAQ,EAAE,MAAM,EAChB,OAAO,EAAG,IAAI,CAAC,WAAW,EAAE,SAAS,GAAG,OAAO,GAAG,MAAM,GAAG,YAAY,CAAC,GACvE,OAAO,CAAC,MAAM,CAAC,CAYjB"}

package/dist/eval/fixtures.js

@@ -0,0 +1,110 @@
+/**
+ * Fixture I/O for `pnpm rudder ai:eval --record` / `--replay` (#A5
+ * Phase 4). Each case writes one JSON file under
+ * `evals/__fixtures__/<suite>/<case>.json` carrying the assistant
+ * turns from a real provider run, normalized into the
+ * {@link AiFakeStep} shape so `--replay` can re-feed them via
+ * `AiFake.respondWithSequence` for zero-API regression tests.
+ *
+ * The fixture format is versioned. Bumping `version` forces a
+ * re-record on stale fixtures rather than silently mis-replaying.
+ */
+import { mkdir, readFile, writeFile } from 'node:fs/promises';
+import path from 'node:path';
+/**
+ * Convert an `AgentResponse` into the assistant-turn `AiFakeStep[]`
+ * sequence that `AiFake.respondWithSequence` expects.
+ *
+ * - Drops user/tool turns — those are framework-generated during a
+ *   replayed run, not provider output.
+ * - Multi-modal assistant content collapses to its concatenated text
+ *   parts (the fake's transport is text-only; image/document parts
+ *   wouldn't replay meaningfully).
+ * - `toolCalls` carry through verbatim so multi-step tool loops
+ *   replay deterministically.
+ */
+export function stepsFromResponse(response) {
+    return response.steps
+        .filter(step => step.message.role === 'assistant')
+        .map(step => {
+        const out = {
+            text: contentToText(step.message.content),
+            finishReason: step.finishReason,
+        };
+        if (step.toolCalls.length > 0)
+            out.toolCalls = step.toolCalls;
+        return out;
+    });
+}
+function contentToText(content) {
+    if (typeof content === 'string')
+        return content;
+    return content.filter(p => p.type === 'text').map(p => p.text).join('');
+}
+// ─── Fixture path conventions ─────────────────────────────
+/**
+ * Default fixtures directory: `<cwd>/evals/__fixtures__`. Override
+ * via the CLI handler's options for tests / non-standard layouts.
+ */
+export function defaultFixturesDir(cwd) {
+    return path.join(cwd, 'evals', '__fixtures__');
+}
+/**
+ * Filesystem-safe slug for `<suite>/<case>` segments. Letters,
+ * digits, dot, dash, underscore pass through; everything else
+ * collapses to `-`. Multiple consecutive `-` collapse to one.
+ *
+ * Pure function; tested directly so suite/case rename diffs stay
+ * predictable across editors.
+ */
+export function slugify(s) {
+    return s.replace(/[^A-Za-z0-9._-]+/g, '-').replace(/^-+|-+$/g, '') || '_';
+}
+export function fixturePath(dir, suite, caseName) {
+    return path.join(dir, slugify(suite), `${slugify(caseName)}.json`);
+}
+// ─── Read / write ─────────────────────────────────────────
+/**
+ * Read a fixture file. Returns `null` when the fixture is missing
+ * (replay falls back to running normally with a clear stderr line).
+ *
+ * Throws on parse / version errors — corruption is not a passing
+ * case and silently ignoring it would mask real regressions.
+ */
+export async function readFixture(dir, suite, caseName) {
+    const file = fixturePath(dir, suite, caseName);
+    let raw;
+    try {
+        raw = await readFile(file, 'utf8');
+    }
+    catch (err) {
+        if (err.code === 'ENOENT')
+            return null;
+        throw err;
+    }
+    const parsed = JSON.parse(raw);
+    if (parsed.version !== 1) {
+        throw new Error(`[RudderJS AI] Fixture ${file} is version ${String(parsed.version)}; expected 1. ` +
+            `Re-record with \`pnpm rudder ai:eval --record\`.`);
+    }
+    return parsed;
+}
+/**
+ * Write a fixture, creating intermediate directories as needed.
+ * Pretty-printed (2-space) so PR diffs remain readable when the
+ * model output evolves.
+ */
+export async function writeFixture(dir, suite, caseName, payload) {
+    const file = fixturePath(dir, suite, caseName);
+    await mkdir(path.dirname(file), { recursive: true });
+    const fixture = {
+        version: 1,
+        suite,
+        case: caseName,
+        recordedAt: new Date().toISOString(),
+        ...payload,
+    };
+    await writeFile(file, `${JSON.stringify(fixture, null, 2)}\n`);
+    return file;
+}
+//# sourceMappingURL=fixtures.js.map

package/dist/eval/fixtures.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"fixtures.js","sourceRoot":"","sources":["../../src/eval/fixtures.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH,OAAO,EAAE,KAAK,EAAE,QAAQ,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAA;AAC7D,OAAO,IAAI,MAAM,WAAW,CAAA;AAc5B;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,iBAAiB,CAAC,QAAuB;IACvD,OAAO,QAAQ,CAAC,KAAK;SAClB,MAAM,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,OAAO,CAAC,IAAI,KAAK,WAAW,CAAC;SACjD,GAAG,CAAC,IAAI,CAAC,EAAE;QACV,MAAM,GAAG,GAAe;YACtB,IAAI,EAAU,aAAa,CAAC,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC;YACjD,YAAY,EAAE,IAAI,CAAC,YAAY;SAChC,CAAA;QACD,IAAI,IAAI,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC;YAAE,GAAG,CAAC,SAAS,GAAG,IAAI,CAAC,SAAS,CAAA;QAC7D,OAAO,GAAG,CAAA;IACZ,CAAC,CAAC,CAAA;AACN,CAAC;AAED,SAAS,aAAa,CAAC,OAA+B;IACpD,IAAI,OAAO,OAAO,KAAK,QAAQ;QAAE,OAAO,OAAO,CAAA;IAC/C,OAAO,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAA;AACzE,CAAC;AAED,6DAA6D;AAE7D;;;GAGG;AACH,MAAM,UAAU,kBAAkB,CAAC,GAAW;IAC5C,OAAO,IAAI,CAAC,IAAI,CAAC,GAAG,EAAE,OAAO,EAAE,cAAc,CAAC,CAAA;AAChD,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,UAAU,OAAO,CAAC,CAAS;IAC/B,OAAO,CAAC,CAAC,OAAO,CAAC,mBAAmB,EAAE,GAAG,CAAC,CAAC,OAAO,CAAC,UAAU,EAAE,EAAE,CAAC,IAAI,GAAG,CAAA;AAC3E,CAAC;AAED,MAAM,UAAU,WAAW,CAAC,GAAW,EAAE,KAAa,EAAE,QAAgB;IACtE,OAAO,IAAI,CAAC,IAAI,CAAC,GAAG,EAAE,OAAO,CAAC,KAAK,CAAC,EAAE,GAAG,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAA;AACpE,CAAC;AAED,6DAA6D;AAE7D;;;;;;GAMG;AACH,MAAM,CAAC,KAAK,UAAU,WAAW,CAC/B,GAAgB,EAChB,KAAgB,EAChB,QAAgB;IAEhB,MAAM,IAAI,GAAG,WAAW,CAAC,GAAG,EAAE,KAAK,EAAE,QAAQ,CAAC,CAAA;IAC9C,IAAI,GAAW,CAAA;IACf,IAAI,CAAC;QACH,GAAG,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC,CAAA;IACpC,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,IAAK,GAA6B,CAAC,IAAI,KAAK,QAAQ;YAAE,OAAO,IAAI,CAAA;QACjE,MAAM,GAAG,CAAA;IACX,CAAC;IACD,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAgB,CAAA;IAC7C,IAAI,MAAM,CAAC,OAAO,KAAK,CAAC,EAAE,CAAC;QACzB,MAAM,IAAI,KAAK,CACb,yBAAyB,IAAI,eAAe,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,gBAAgB;YAClF,kDAAkD,CACnD,CAAA;IACH,CAAC;IACD,OAAO,MAAM,CAAA;AACf,CAAC;AAED;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,YAAY,CAChC,GAAgB,EAChB,KAAgB,EAChB,QAAgB,EAChB,OAAwE;IAExE,MAAM,IAAI,GAAG,WAAW,CAAC,GAAG,EAAE,KAAK,EAAE,QAAQ,CAAC,CAAA;IAC9C,MAAM,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAA;IACpD,MAAM,OAAO,GAAgB;QAC3B,OAAO,EAAK,CAAC;QACb,KAAK;QACL,IAAI,EAAQ,QAAQ;QACpB,UAAU,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;QACpC,GAAG,OAAO;KACX,CAAA;IACD,MAAM,SAAS,CAAC,IAAI,EAAE,GAAG,IAAI,CAAC,SAAS,CAAC,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,CAAA;IAC9D,OAAO,IAAI,CAAA;AACb,CAAC"}

package/dist/eval/html-reporter.d.ts

@@ -0,0 +1,25 @@
+/**
+ * Self-contained HTML reporter for `pnpm rudder ai:eval --html`
+ * (#A5 Phase 5). Renders one or more {@link SuiteReport}s into a
+ * single HTML string with inline styles and minimal vanilla JS for
+ * row expand/collapse — no framework deps, no external assets, safe
+ * to paste into a PR comment or open offline.
+ *
+ * Defensive HTML-escape on every piece of user content (suite name,
+ * case name, input, response, metadata). Long responses get a
+ * `<pre>` block with `white-space: pre-wrap` so output stays
+ * scannable without a horizontal scroll.
+ */
+import type { SuiteReport } from './index.js';
+export interface HtmlReportOptions {
+    /** Document `<title>`. Defaults to `"Eval Report"`. */
+    title?: string;
+    /** ISO timestamp shown in the header. Defaults to `new Date().toISOString()`. */
+    generatedAt?: string;
+}
+/**
+ * Render an array of {@link SuiteReport}s as a single self-contained
+ * HTML document.
+ */
+export declare function reportHtml(reports: SuiteReport[], opts?: HtmlReportOptions): string;
+//# sourceMappingURL=html-reporter.d.ts.map

package/dist/eval/html-reporter.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"html-reporter.d.ts","sourceRoot":"","sources":["../../src/eval/html-reporter.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH,OAAO,KAAK,EAAE,WAAW,EAAgB,MAAM,YAAY,CAAA;AAE3D,MAAM,WAAW,iBAAiB;IAChC,uDAAuD;IACvD,KAAK,CAAC,EAAQ,MAAM,CAAA;IACpB,iFAAiF;IACjF,WAAW,CAAC,EAAE,MAAM,CAAA;CACrB;AAED;;;GAGG;AACH,wBAAgB,UAAU,CAAC,OAAO,EAAE,WAAW,EAAE,EAAE,IAAI,GAAE,iBAAsB,GAAG,MAAM,CA6CvF"}

package/dist/eval/html-reporter.js

@@ -0,0 +1,209 @@
+/**
+ * Self-contained HTML reporter for `pnpm rudder ai:eval --html`
+ * (#A5 Phase 5). Renders one or more {@link SuiteReport}s into a
+ * single HTML string with inline styles and minimal vanilla JS for
+ * row expand/collapse — no framework deps, no external assets, safe
+ * to paste into a PR comment or open offline.
+ *
+ * Defensive HTML-escape on every piece of user content (suite name,
+ * case name, input, response, metadata). Long responses get a
+ * `<pre>` block with `white-space: pre-wrap` so output stays
+ * scannable without a horizontal scroll.
+ */
+/**
+ * Render an array of {@link SuiteReport}s as a single self-contained
+ * HTML document.
+ */
+export function reportHtml(reports, opts = {}) {
+    const title = opts.title ?? 'Eval Report';
+    const generatedAt = opts.generatedAt ?? new Date().toISOString();
+    const totals = reports.reduce((a, r) => ({
+        cases: a.cases + r.cases.length,
+        passed: a.passed + r.passed,
+        failed: a.failed + r.failed,
+        skipped: a.skipped + r.skipped,
+        cost: a.cost + r.cost,
+        tokens: a.tokens + r.tokens,
+        duration: a.duration + r.duration,
+    }), { cases: 0, passed: 0, failed: 0, skipped: 0, cost: 0, tokens: 0, duration: 0 });
+    const passRate = totals.cases > 0 ? Math.round((totals.passed / totals.cases) * 100) : 0;
+    return `<!DOCTYPE html>
+<html lang="en">
+<head>
+<meta charset="UTF-8">
+<meta name="viewport" content="width=device-width,initial-scale=1">
+<title>${escapeHtml(title)}</title>
+<style>${STYLE}</style>
+</head>
+<body>
+<header class="page-header">
+  <h1>${escapeHtml(title)}</h1>
+  <div class="meta">
+    Generated ${escapeHtml(generatedAt)} ·
+    ${reports.length} suite${plural(reports.length)} ·
+    ${totals.cases} case${plural(totals.cases)} ·
+    <strong class="${passRate === 100 ? 'ok' : passRate >= 80 ? 'warn' : 'bad'}">${passRate}% pass</strong> ·
+    ${formatCost(totals.cost)} ·
+    ${totals.tokens.toLocaleString()} tokens ·
+    ${formatMs(totals.duration)}
+  </div>
+</header>
+${reports.map(renderSuite).join('\n')}
+<script>${SCRIPT}</script>
+</body>
+</html>
+`;
+}
+function renderSuite(r) {
+    const passRate = r.cases.length > 0 ? Math.round((r.passed / r.cases.length) * 100) : 0;
+    return `<section class="suite">
+  <header class="suite-header">
+    <h2>${escapeHtml(r.suite)}</h2>
+    <div class="suite-stats">
+      <span class="${passRate === 100 ? 'ok' : passRate >= 80 ? 'warn' : 'bad'}">${r.passed}/${r.cases.length} passed</span>
+      ${r.skipped > 0 ? `· <span class="muted">${r.skipped} skipped</span>` : ''}
+      · ${formatCost(r.cost)}
+      · ${r.tokens.toLocaleString()} tokens
+      · ${formatMs(r.duration)}
+    </div>
+    ${renderMetadata(r.metadata)}
+  </header>
+  <table class="cases">
+    <thead>
+      <tr>
+        <th>Case</th>
+        <th>Status</th>
+        <th class="num">Tokens</th>
+        <th class="num">Cost</th>
+        <th class="num">Duration</th>
+      </tr>
+    </thead>
+    <tbody>
+${r.cases.map(renderCase).join('\n')}
+    </tbody>
+  </table>
+</section>`;
+}
+function renderCase(c) {
+    const glyph = c.status === 'passed' ? '✓' : c.status === 'failed' ? '✗' : '○';
+    const responseBlock = c.responseText !== undefined
+        ? `<h4>Response</h4><pre>${escapeHtml(c.responseText)}</pre>`
+        : '<h4>Response</h4><pre class="muted">&lt;no response — agent threw or skipped&gt;</pre>';
+    const reasonBlock = (c.metric?.reason ?? c.reason)
+        ? `<h4>Reason</h4><pre>${escapeHtml(c.metric?.reason ?? c.reason)}</pre>`
+        : '';
+    const scoreBlock = c.metric?.score !== undefined
+        ? `<h4>Score</h4><pre>${c.metric.score.toFixed(3)}</pre>`
+        : '';
+    return `      <tr class="case ${c.status}" tabindex="0" aria-expanded="false">
+        <td><span class="glyph">${glyph}</span> ${escapeHtml(c.name)}</td>
+        <td><span class="badge ${c.status}">${c.status}</span></td>
+        <td class="num">${c.tokens.toLocaleString()}</td>
+        <td class="num">${formatCost(c.cost)}</td>
+        <td class="num">${formatMs(c.duration)}</td>
+      </tr>
+      <tr class="case-detail" hidden>
+        <td colspan="5">
+          <h4>Input</h4>
+          <pre>${escapeHtml(c.input)}</pre>
+          ${responseBlock}
+          ${scoreBlock}
+          ${reasonBlock}
+        </td>
+      </tr>`;
+}
+function renderMetadata(meta) {
+    if (!meta)
+        return '';
+    const rows = Object.entries(meta).filter(([, v]) => v !== undefined && v !== '');
+    if (rows.length === 0)
+        return '';
+    return `<dl class="metadata">${rows.map(([k, v]) => `<dt>${escapeHtml(formatLabel(k))}</dt><dd>${escapeHtml(v)}</dd>`).join('')}</dl>`;
+}
+function formatLabel(key) {
+    // camelCase → Title Case for the well-known keys; pass others through.
+    if (key === 'lastReviewed')
+        return 'Last reviewed';
+    return key.charAt(0).toUpperCase() + key.slice(1);
+}
+// ─── HTML escape (no external dep) ───────────────────────
+const ESCAPE_MAP = {
+    '&': '&amp;',
+    '<': '&lt;',
+    '>': '&gt;',
+    '"': '&quot;',
+    "'": '&#39;',
+};
+function escapeHtml(s) {
+    return s.replace(/[&<>"']/g, ch => ESCAPE_MAP[ch]);
+}
+function plural(n) {
+    return n === 1 ? '' : 's';
+}
+function formatMs(ms) {
+    if (ms < 1000)
+        return `${Math.round(ms)}ms`;
+    return `${(ms / 1000).toFixed(1)}s`;
+}
+function formatCost(usd) {
+    if (usd === 0)
+        return '$0.000';
+    if (usd < 0.001)
+        return '<$0.001';
+    return `$${usd.toFixed(3)}`;
+}
+// ─── Inline assets ────────────────────────────────────────
+const STYLE = `
+:root { color-scheme: light dark; --fg: #1a1a1a; --bg: #fff; --muted: #6a6a6a; --border: #e2e2e2; --row-hover: #f7f7f7; --ok: #1a7f37; --warn: #b08800; --bad: #b91c1c; --pre-bg: #f6f8fa; }
+@media (prefers-color-scheme: dark) {
+  :root { --fg: #e6e6e6; --bg: #0d1117; --muted: #8a8a8a; --border: #30363d; --row-hover: #161b22; --ok: #3fb950; --warn: #d29922; --bad: #f85149; --pre-bg: #161b22; }
+}
+* { box-sizing: border-box }
+body { font: 14px/1.5 -apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif; color: var(--fg); background: var(--bg); margin: 0; padding: 24px; max-width: 1100px; margin-inline: auto }
+h1 { margin: 0 0 4px; font-size: 24px }
+h2 { margin: 0 0 4px; font-size: 18px }
+h4 { margin: 12px 0 4px; font-size: 12px; text-transform: uppercase; letter-spacing: .04em; color: var(--muted) }
+pre { background: var(--pre-bg); border: 1px solid var(--border); border-radius: 6px; padding: 8px 12px; overflow-x: auto; white-space: pre-wrap; word-wrap: break-word; font: 12px/1.45 ui-monospace,SFMono-Regular,Menlo,monospace; margin: 0 }
+.page-header { border-bottom: 1px solid var(--border); padding-bottom: 12px; margin-bottom: 20px }
+.page-header .meta { color: var(--muted); font-size: 13px }
+.suite { margin-bottom: 28px }
+.suite-header h2 { display: inline }
+.suite-stats { color: var(--muted); font-size: 13px; margin-top: 4px }
+.metadata { display: grid; grid-template-columns: max-content 1fr; gap: 4px 12px; margin: 8px 0 0; font-size: 13px }
+.metadata dt { color: var(--muted); font-weight: normal }
+.metadata dd { margin: 0 }
+.cases { width: 100%; border-collapse: collapse; margin-top: 12px; border: 1px solid var(--border); border-radius: 6px; overflow: hidden }
+.cases th, .cases td { padding: 8px 12px; text-align: left; border-bottom: 1px solid var(--border) }
+.cases th { background: var(--pre-bg); font-weight: 600; font-size: 12px; text-transform: uppercase; letter-spacing: .04em; color: var(--muted) }
+.cases tr:last-child td { border-bottom: none }
+.cases tr.case { cursor: pointer; user-select: none }
+.cases tr.case:hover { background: var(--row-hover) }
+.cases tr.case:focus { outline: 2px solid var(--warn); outline-offset: -2px }
+.cases tr.case-detail td { background: var(--pre-bg) }
+.cases td.num { text-align: right; font-variant-numeric: tabular-nums }
+.glyph { display: inline-block; width: 14px; font-weight: bold }
+.cases tr.case.passed .glyph { color: var(--ok) }
+.cases tr.case.failed .glyph { color: var(--bad) }
+.cases tr.case.skipped .glyph { color: var(--muted) }
+.badge { display: inline-block; padding: 1px 8px; border-radius: 999px; font-size: 11px; font-weight: 600; text-transform: uppercase; letter-spacing: .04em }
+.badge.passed { background: rgba(63,185,80,.15); color: var(--ok) }
+.badge.failed { background: rgba(248,81,73,.15); color: var(--bad) }
+.badge.skipped { background: rgba(138,138,138,.15); color: var(--muted) }
+.ok { color: var(--ok) } .warn { color: var(--warn) } .bad { color: var(--bad) } .muted { color: var(--muted) }
+`.trim();
+const SCRIPT = `
+document.querySelectorAll('tr.case').forEach(function(row) {
+  function toggle() {
+    var detail = row.nextElementSibling;
+    if (!detail || !detail.classList.contains('case-detail')) return;
+    var open = !detail.hidden;
+    detail.hidden = open;
+    row.setAttribute('aria-expanded', String(!open));
+  }
+  row.addEventListener('click', toggle);
+  row.addEventListener('keydown', function(e) {
+    if (e.key === 'Enter' || e.key === ' ') { e.preventDefault(); toggle(); }
+  });
+});
+`.trim();
+//# sourceMappingURL=html-reporter.js.map

package/dist/eval/html-reporter.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"html-reporter.js","sourceRoot":"","sources":["../../src/eval/html-reporter.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAWH;;;GAGG;AACH,MAAM,UAAU,UAAU,CAAC,OAAsB,EAAE,OAA0B,EAAE;IAC7E,MAAM,KAAK,GAAS,IAAI,CAAC,KAAK,IAAI,aAAa,CAAA;IAC/C,MAAM,WAAW,GAAG,IAAI,CAAC,WAAW,IAAI,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAA;IAEhE,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAC3B,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC;QACT,KAAK,EAAK,CAAC,CAAC,KAAK,GAAM,CAAC,CAAC,KAAK,CAAC,MAAM;QACrC,MAAM,EAAI,CAAC,CAAC,MAAM,GAAK,CAAC,CAAC,MAAM;QAC/B,MAAM,EAAI,CAAC,CAAC,MAAM,GAAK,CAAC,CAAC,MAAM;QAC/B,OAAO,EAAG,CAAC,CAAC,OAAO,GAAI,CAAC,CAAC,OAAO;QAChC,IAAI,EAAM,CAAC,CAAC,IAAI,GAAO,CAAC,CAAC,IAAI;QAC7B,MAAM,EAAI,CAAC,CAAC,MAAM,GAAK,CAAC,CAAC,MAAM;QAC/B,QAAQ,EAAE,CAAC,CAAC,QAAQ,GAAG,CAAC,CAAC,QAAQ;KAClC,CAAC,EACF,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,QAAQ,EAAE,CAAC,EAAE,CAChF,CAAA;IAED,MAAM,QAAQ,GAAG,MAAM,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,MAAM,CAAC,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAA;IAExF,OAAO;;;;;SAKA,UAAU,CAAC,KAAK,CAAC;SACjB,KAAK;;;;QAIN,UAAU,CAAC,KAAK,CAAC;;gBAET,UAAU,CAAC,WAAW,CAAC;MACjC,OAAO,CAAC,MAAM,SAAS,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC;MAC7C,MAAM,CAAC,KAAK,QAAQ,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC;qBACzB,QAAQ,KAAK,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,QAAQ,IAAI,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,KAAK,KAAK,QAAQ;MACrF,UAAU,CAAC,MAAM,CAAC,IAAI,CAAC;MACvB,MAAM,CAAC,MAAM,CAAC,cAAc,EAAE;MAC9B,QAAQ,CAAC,MAAM,CAAC,QAAQ,CAAC;;;EAG7B,OAAO,CAAC,GAAG,CAAC,WAAW,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC;UAC3B,MAAM;;;CAGf,CAAA;AACD,CAAC;AAED,SAAS,WAAW,CAAC,CAAc;IACjC,MAAM,QAAQ,GAAG,CAAC,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAA;IACvF,OAAO;;UAEC,UAAU,CAAC,CAAC,CAAC,KAAK,CAAC;;qBAER,QAAQ,KAAK,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,QAAQ,IAAI,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,KAAK,KAAK,CAAC,CAAC,MAAM,IAAI,CAAC,CAAC,KAAK,CAAC,MAAM;QACrG,CAAC,CAAC,OAAO,GAAG,CAAC,CAAC,CAAC,CAAC,yBAAyB,CAAC,CAAC,OAAO,iBAAiB,CAAC,CAAC,CAAC,EAAE;UACtE,UAAU,CAAC,CAAC,CAAC,IAAI,CAAC;UAClB,CAAC,CAAC,MAAM,CAAC,cAAc,EAAE;UACzB,QAAQ,CAAC,CAAC,CAAC,QAAQ,CAAC;;MAExB,cAAc,CAAC,CAAC,CAAC,QAAQ,CAAC;;;;;;;;;;;;;EAa9B,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC;;;WAGzB,CAAA;AACX,CAAC;AAED,SAAS,UAAU,CAAC,CAA+B;IACjD,MAAM,KAAK,GAAG,CAAC,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAA;IAC7E,MAAM,aAAa,GAAG,CAAC,CAAC,YAAY,KAAK,SAAS;QAChD,CAAC,CAAC,yBAAyB,UAAU,CAAC,CAAC,CAAC,YAAY,CAAC,QAAQ;QAC7D,CAAC,CAAC,wFAAwF,CAAA;IAC5F,MAAM,WAAW,GAAG,CAAC,CAAC,CAAC,MAAM,EAAE,MAAM,IAAI,CAAC,CAAC,MAAM,CAAC;QAChD,CAAC,CAAC,uBAAuB,UAAU,CAAC,CAAC,CAAC,MAAM,EAAE,MAAM,IAAI,CAAC,CAAC,MAAO,CAAC,QAAQ;QAC1E,CAAC,CAAC,EAAE,CAAA;IACN,MAAM,UAAU,GAAG,CAAC,CAAC,MAAM,EAAE,KAAK,KAAK,SAAS;QAC9C,CAAC,CAAC,sBAAsB,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,QAAQ;QACzD,CAAC,CAAC,EAAE,CAAA;IACN,OAAO,yBAAyB,CAAC,CAAC,MAAM;kCACR,KAAK,WAAW,UAAU,CAAC,CAAC,CAAC,IAAI,CAAC;iCACnC,CAAC,CAAC,MAAM,KAAK,CAAC,CAAC,MAAM;0BAC5B,CAAC,CAAC,MAAM,CAAC,cAAc,EAAE;0BACzB,UAAU,CAAC,CAAC,CAAC,IAAI,CAAC;0BAClB,QAAQ,CAAC,CAAC,CAAC,QAAQ,CAAC;;;;;iBAK7B,UAAU,CAAC,CAAC,CAAC,KAAK,CAAC;YACxB,aAAa;YACb,UAAU;YACV,WAAW;;YAEX,CAAA;AACZ,CAAC;AAED,SAAS,cAAc,CAAC,IAA8B;IACpD,IAAI,CAAC,IAAI;QAAE,OAAO,EAAE,CAAA;IACpB,MAAM,IAAI,GAAG,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,KAAK,SAAS,IAAI,CAAC,KAAK,EAAE,CAAC,CAAA;IAChF,IAAI,IAAI,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAA;IAChC,OAAO,wBACL,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,UAAU,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,YAAY,UAAU,CAAC,CAAE,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,EAAE,CAClG,OAAO,CAAA;AACT,CAAC;AAED,SAAS,WAAW,CAAC,GAAW;IAC9B,uEAAuE;IACvE,IAAI,GAAG,KAAK,cAAc;QAAE,OAAO,eAAe,CAAA;IAClD,OAAO,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,WAAW,EAAE,GAAG,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,CAAA;AACnD,CAAC;AAED,4DAA4D;AAE5D,MAAM,UAAU,GAA2B;IACzC,GAAG,EAAG,OAAO;IACb,GAAG,EAAG,MAAM;IACZ,GAAG,EAAG,MAAM;IACZ,GAAG,EAAG,QAAQ;IACd,GAAG,EAAG,OAAO;CACd,CAAA;AACD,SAAS,UAAU,CAAC,CAAS;IAC3B,OAAO,CAAC,CAAC,OAAO,CAAC,UAAU,EAAE,EAAE,CAAC,EAAE,CAAC,UAAU,CAAC,EAAE,CAAE,CAAC,CAAA;AACrD,CAAC;AAED,SAAS,MAAM,CAAC,CAAS;IACvB,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,CAAA;AAC3B,CAAC;AAED,SAAS,QAAQ,CAAC,EAAU;IAC1B,IAAI,EAAE,GAAG,IAAI;QAAE,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,IAAI,CAAA;IAC3C,OAAO,GAAG,CAAC,EAAE,GAAG,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAA;AACrC,CAAC;AAED,SAAS,UAAU,CAAC,GAAW;IAC7B,IAAI,GAAG,KAAK,CAAC;QAAM,OAAO,QAAQ,CAAA;IAClC,IAAI,GAAG,GAAG,KAAK;QAAI,OAAO,SAAS,CAAA;IACnC,OAAO,IAAI,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAA;AAC7B,CAAC;AAED,6DAA6D;AAE7D,MAAM,KAAK,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CAqCb,CAAC,IAAI,EAAE,CAAA;AAER,MAAM,MAAM,GAAG;;;;;;;;;;;;;;CAcd,CAAC,IAAI,EAAE,CAAA"}