@opensearch-project/agent-health 0.4.0 → 0.6.0

package/lib/dist/lib/testCases/loader.js

@@ -0,0 +1,179 @@
+/*
+ * Copyright OpenSearch Contributors
+ * SPDX-License-Identifier: Apache-2.0
+ */
+import { createHash } from 'crypto';
+import { readFileSync } from 'fs';
+import { resolve, dirname } from 'path';
+import { pathToFileURL } from 'url';
+import { createRequire, Module as NodeModule } from 'module';
+import { setActiveFile, getRegisteredTests, getRegisteredHooks, clearRegistry, } from './define.js';
+import { getAuthoringSurface } from './authoringSurface.js';
+const CODE_EXTENSIONS = ['.ts', '.js', '.mjs'];
+export function isCodeFile(filename) {
+    const lower = filename.toLowerCase();
+    return CODE_EXTENSIONS.some(ext => lower.endsWith(ext));
+}
+// Re-exported so CLI/server importers keep a single import path
+// (`@/lib/testCases/loader`) for both the loader and the language
+// detector, while the browser-side EvalSourceCodeView component imports
+// the same isomorphic implementation directly from `@/lib/utils` (this
+// module pulls in `fs`/`module`, which are Node-only and unsafe to bundle
+// into the browser).
+export { detectSourceLanguage } from '../../lib/utils.js';
+export function computeTestCaseHash(tc, fileSource) {
+    const content = JSON.stringify({
+        name: tc.name,
+        prompt: tc.options.prompt,
+        context: tc.options.context,
+        labels: tc.options.labels,
+        description: tc.options.description,
+        // Include the new expected* fields so editing them invalidates the
+        // sourceHash and the upsert path picks up the change. Without this,
+        // a user adding `expectedOutcomes: [...]` to an existing test would
+        // see the test case stay on its old version and the new outcomes
+        // would never reach storage.
+        expectedOutcomes: tc.options.expectedOutcomes,
+        expectedTrajectory: tc.options.expectedTrajectory,
+        // Fold in the WHOLE file's raw text (optional -- omitted by callers that
+        // don't have it, e.g. existing unit tests exercising this function in
+        // isolation). Without this, editing ONLY the evaluate() body, a helper
+        // function, an import, or even just a comment would leave every
+        // options-derived field above unchanged, `sourceHash` would stay put,
+        // `bulkUpsert` would classify the row as unchanged, and the persisted
+        // `sourceCode` -- the entire point of the eval-source viewer -- would
+        // silently go stale relative to the real file on disk.
+        fileSource,
+    });
+    return createHash('sha256').update(content).digest('hex');
+}
+export async function loadTestCasesFromModule(filePath) {
+    const absPath = resolve(filePath);
+    // Read the raw file text up front, before any execution path below --
+    // both the .js (already needed its own read for the CJS wrapper) and the
+    // .ts/.mjs (dynamic `import()`) paths reuse this single read. For .js this
+    // read IS the executed source (the CJS wrapper below runs this exact
+    // string). For .ts/.mjs, `import()` reads the file independently via
+    // Node's own module loader a few lines later -- in the extremely narrow
+    // window where the file is edited between this read and that import
+    // resolving, `fileSource` could theoretically diverge from what actually
+    // ran. That's an accepted, practically negligible risk (same file, same
+    // synchronous-ish call, no network hop) rather than a guarantee.
+    const fileSource = readFileSync(absPath, 'utf-8');
+    // Clear any prior registration for this file and set it as active
+    clearRegistry(absPath);
+    setActiveFile(absPath);
+    let module;
+    if (absPath.endsWith('.js')) {
+        // For CJS .js files, execute in a fresh context with our test() function
+        // injected. This avoids module caching issues across multiple loads.
+        const code = fileSource;
+        const fileDir = dirname(absPath);
+        // NOTE: do not call require('module') here — this file is bundled as ESM
+        // by esbuild for the server, and Dynamic require of built-ins is unsupported
+        // in ESM output. Use the statically-imported NodeModule instead.
+        const Module = NodeModule;
+        const m = new Module(absPath);
+        m.filename = absPath;
+        m.paths = Module._nodeModulePaths(fileDir);
+        // Provide a require function scoped to the file's directory, but override
+        // any require of the define module to return our own instance
+        const fileRequire = createRequire(absPath);
+        // Match if `id` looks like our own define module before doing any
+        // filesystem resolution. Required because:
+        //   1) the loader is bundled as ESM by esbuild, so synthetic CJS
+        //      `require.resolve` is unavailable here, and
+        //   2) `lib/testCases/define.ts` is not compiled to a .js sibling, so
+        //      `fileRequire.resolve('../../lib/testCases/define')` from a
+        //      fixture would throw MODULE_NOT_FOUND.
+        const isDefineId = (id) => {
+            const normalized = id.replace(/\\/g, '/').replace(/\.js$/, '');
+            return (normalized === 'lib/testCases/define' ||
+                normalized.endsWith('/lib/testCases/define'));
+        };
+        // Intercept the published package name as well — fixtures generally
+        // do `require('@opensearch-project/agent-health')` and Node's resolver
+        // would otherwise hit the package's exports map (which only exposes
+        // 'import' for ESM consumers).
+        const isPackageName = (id) => id === '@opensearch-project/agent-health' ||
+            id === '@opensearch/agent-health' ||
+            id === 'agent-health';
+        // The object handed back when a CJS eval file requires the SDK. Single
+        // source of truth shared with the package exports (see authoringSurface)
+        // so `.js` and `.ts`/`.mjs` files see the SAME surface — no drift (#232).
+        const sdkExports = getAuthoringSurface();
+        const wrappedRequire = (id) => {
+            if (isDefineId(id) || isPackageName(id)) {
+                return sdkExports;
+            }
+            // Fall back to a resolution-based check so that fixtures requiring the
+            // SDK by absolute path or via the published package name still hand back
+            // our test() registrar (and therefore land in our file-scoped registry).
+            try {
+                const resolved = fileRequire.resolve(id);
+                const normalized = resolved.replace(/\\/g, '/');
+                if (normalized.endsWith('/lib/testCases/define.js') ||
+                    normalized.endsWith('/lib/testCases/define')) {
+                    return sdkExports;
+                }
+            }
+            catch {
+                // ignore unresolvable IDs — fileRequire below will surface the error
+            }
+            return fileRequire(id);
+        };
+        wrappedRequire.resolve = fileRequire.resolve;
+        const wrapper = `(function(exports, require, module, __filename, __dirname) { ${code}\n});`;
+        const compiledFn = eval(wrapper);
+        compiledFn(m.exports, wrappedRequire, m, absPath, fileDir);
+        module = m.exports;
+    }
+    else {
+        // For .ts and .mjs files, use dynamic import
+        try {
+            const fileUrl = pathToFileURL(absPath).href;
+            module = await import(fileUrl);
+        }
+        catch (err) {
+            if (err.code === 'ERR_UNKNOWN_FILE_EXTENSION' && absPath.endsWith('.ts')) {
+                throw new Error(`Cannot import TypeScript file: ${filePath}\n` +
+                    'Install tsx as a dependency: npm install tsx\n' +
+                    'Or pre-compile .eval.ts to .eval.js before running.');
+            }
+            throw new Error(`Failed to import module: ${filePath}\n${err.message}`);
+        }
+    }
+    // Tests come exclusively from test() / describe() registration calls
+    // (Playwright-style). Legacy default-export array form was removed when
+    // we dropped defineTestCases() — the SDK is experimental and the API
+    // shape is intentionally narrow.
+    const testCases = getRegisteredTests(absPath);
+    if (testCases.length === 0) {
+        throw new Error(`Module ${filePath} has no test cases. Register tests with test(name, options?, body) inside the file.`);
+    }
+    // Compute per-test-case hash. Includes fileSource (see
+    // computeTestCaseHash) so ANY change to the file -- not just the
+    // options fields already tracked -- invalidates the hash and triggers a
+    // re-persist, keeping the stored sourceCode in sync with disk.
+    const loaded = testCases.map(tc => ({
+        ...tc,
+        hash: computeTestCaseHash(tc, fileSource),
+    }));
+    // Derive describe-group → test names mapping. Tests outside any
+    // describe() have benchmarkPath=undefined and are excluded; the CLI/
+    // server adds them to the file-default benchmark separately.
+    const benchmarks = new Map();
+    for (const tc of loaded) {
+        if (tc.benchmarkPath) {
+            const list = benchmarks.get(tc.benchmarkPath) ?? [];
+            list.push(tc.name);
+            benchmarks.set(tc.benchmarkPath, list);
+        }
+    }
+    // Collect every hook the file registered. Empty when no
+    // beforeEach/afterEach/beforeAll/afterAll were called — the orchestrator
+    // is a no-op in that case so existing tests are unaffected.
+    const hooks = getRegisteredHooks(absPath);
+    return { testCases: loaded, filePath: absPath, benchmarks, hooks, fileSource };
+}
+//# sourceMappingURL=loader.js.map

package/lib/dist/lib/testCases/loader.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"loader.js","sourceRoot":"","sources":["../../../testCases/loader.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,EAAE,UAAU,EAAE,MAAM,QAAQ,CAAC;AACpC,OAAO,EAAE,YAAY,EAAE,MAAM,IAAI,CAAC;AAClC,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,MAAM,CAAC;AACxC,OAAO,EAAE,aAAa,EAAE,MAAM,KAAK,CAAC;AACpC,OAAO,EAAE,aAAa,EAAE,MAAM,IAAI,UAAU,EAAE,MAAM,QAAQ,CAAC;AAE7D,OAAO,EACL,aAAa,EACb,kBAAkB,EAClB,kBAAkB,EAClB,aAAa,GACd,MAAM,aAAa,CAAC;AACrB,OAAO,EAAE,mBAAmB,EAAE,MAAM,uBAAuB,CAAC;AAE5D,MAAM,eAAe,GAAG,CAAC,KAAK,EAAE,KAAK,EAAE,MAAM,CAAC,CAAC;AAE/C,MAAM,UAAU,UAAU,CAAC,QAAgB;IACzC,MAAM,KAAK,GAAG,QAAQ,CAAC,WAAW,EAAE,CAAC;IACrC,OAAO,eAAe,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC;AAC1D,CAAC;AAED,gEAAgE;AAChE,kEAAkE;AAClE,wEAAwE;AACxE,uEAAuE;AACvE,0EAA0E;AAC1E,qBAAqB;AACrB,OAAO,EAAE,oBAAoB,EAAE,MAAM,gBAAgB,CAAC;AAEtD,MAAM,UAAU,mBAAmB,CAAC,EAAgB,EAAE,UAAmB;IACvE,MAAM,OAAO,GAAG,IAAI,CAAC,SAAS,CAAC;QAC7B,IAAI,EAAE,EAAE,CAAC,IAAI;QACb,MAAM,EAAE,EAAE,CAAC,OAAO,CAAC,MAAM;QACzB,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,OAAO;QAC3B,MAAM,EAAE,EAAE,CAAC,OAAO,CAAC,MAAM;QACzB,WAAW,EAAE,EAAE,CAAC,OAAO,CAAC,WAAW;QACnC,mEAAmE;QACnE,oEAAoE;QACpE,oEAAoE;QACpE,iEAAiE;QACjE,6BAA6B;QAC7B,gBAAgB,EAAE,EAAE,CAAC,OAAO,CAAC,gBAAgB;QAC7C,kBAAkB,EAAE,EAAE,CAAC,OAAO,CAAC,kBAAkB;QACjD,yEAAyE;QACzE,sEAAsE;QACtE,uEAAuE;QACvE,gEAAgE;QAChE,sEAAsE;QACtE,sEAAsE;QACtE,sEAAsE;QACtE,uDAAuD;QACvD,UAAU;KACX,CAAC,CAAC;IACH,OAAO,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;AAC5D,CAAC;AAkCD,MAAM,CAAC,KAAK,UAAU,uBAAuB,CAAC,QAAgB;IAC5D,MAAM,OAAO,GAAG,OAAO,CAAC,QAAQ,CAAC,CAAC;IAElC,sEAAsE;IACtE,yEAAyE;IACzE,2EAA2E;IAC3E,qEAAqE;IACrE,qEAAqE;IACrE,wEAAwE;IACxE,oEAAoE;IACpE,yEAAyE;IACzE,wEAAwE;IACxE,iEAAiE;IACjE,MAAM,UAAU,GAAG,YAAY,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;IAElD,kEAAkE;IAClE,aAAa,CAAC,OAAO,CAAC,CAAC;IACvB,aAAa,CAAC,OAAO,CAAC,CAAC;IAEvB,IAAI,MAAW,CAAC;IAEhB,IAAI,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC;QAC5B,yEAAyE;QACzE,qEAAqE;QACrE,MAAM,IAAI,GAAG,UAAU,CAAC;QACxB,MAAM,OAAO,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC;QACjC,yEAAyE;QACzE,6EAA6E;QAC7E,iEAAiE;QACjE,MAAM,MAAM,GAAG,UAAqC,CAAC;QACrD,MAAM,CAAC,GAAG,IAAK,MAAc,CAAC,OAAO,CAAC,CAAC;QACvC,CAAC,CAAC,QAAQ,GAAG,OAAO,CAAC;QACrB,CAAC,CAAC,KAAK,GAAI,MAAc,CAAC,gBAAgB,CAAC,OAAO,CAAC,CAAC;QACpD,0EAA0E;QAC1E,8DAA8D;QAC9D,MAAM,WAAW,GAAG,aAAa,CAAC,OAAO,CAAC,CAAC;QAC3C,kEAAkE;QAClE,2CAA2C;QAC3C,iEAAiE;QACjE,kDAAkD;QAClD,sEAAsE;QACtE,kEAAkE;QAClE,6CAA6C;QAC7C,MAAM,UAAU,GAAG,CAAC,EAAU,EAAE,EAAE;YAChC,MAAM,UAAU,GAAG,EAAE,CAAC,OAAO,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;YAC/D,OAAO,CACL,UAAU,KAAK,sBAAsB;gBACrC,UAAU,CAAC,QAAQ,CAAC,uBAAuB,CAAC,CAC7C,CAAC;QACJ,CAAC,CAAC;QACF,oEAAoE;QACpE,uEAAuE;QACvE,oEAAoE;QACpE,+BAA+B;QAC/B,MAAM,aAAa,GAAG,CAAC,EAAU,EAAE,EAAE,CACnC,EAAE,KAAK,kCAAkC;YACzC,EAAE,KAAK,0BAA0B;YACjC,EAAE,KAAK,cAAc,CAAC;QACxB,uEAAuE;QACvE,yEAAyE;QACzE,0EAA0E;QAC1E,MAAM,UAAU,GAAG,mBAAmB,EAAE,CAAC;QACzC,MAAM,cAAc,GAAG,CAAC,EAAU,EAAE,EAAE;YACpC,IAAI,UAAU,CAAC,EAAE,CAAC,IAAI,aAAa,CAAC,EAAE,CAAC,EAAE,CAAC;gBACxC,OAAO,UAAU,CAAC;YACpB,CAAC;YACD,uEAAuE;YACvE,yEAAyE;YACzE,yEAAyE;YACzE,IAAI,CAAC;gBACH,MAAM,QAAQ,GAAG,WAAW,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC;gBACzC,MAAM,UAAU,GAAG,QAAQ,CAAC,OAAO,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;gBAChD,IACE,UAAU,CAAC,QAAQ,CAAC,0BAA0B,CAAC;oBAC/C,UAAU,CAAC,QAAQ,CAAC,uBAAuB,CAAC,EAC5C,CAAC;oBACD,OAAO,UAAU,CAAC;gBACpB,CAAC;YACH,CAAC;YAAC,MAAM,CAAC;gBACP,qEAAqE;YACvE,CAAC;YACD,OAAO,WAAW,CAAC,EAAE,CAAC,CAAC;QACzB,CAAC,CAAC;QACD,cAAsB,CAAC,OAAO,GAAG,WAAW,CAAC,OAAO,CAAC;QAEtD,MAAM,OAAO,GAAG,gEAAgE,IAAI,OAAO,CAAC;QAC5F,MAAM,UAAU,GAAG,IAAI,CAAC,OAAO,CAAC,CAAC;QACjC,UAAU,CAAC,CAAC,CAAC,OAAO,EAAE,cAAc,EAAE,CAAC,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;QAC3D,MAAM,GAAG,CAAC,CAAC,OAAO,CAAC;IACrB,CAAC;SAAM,CAAC;QACN,6CAA6C;QAC7C,IAAI,CAAC;YACH,MAAM,OAAO,GAAG,aAAa,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC;YAC5C,MAAM,GAAG,MAAM,MAAM,CAAC,OAAO,CAAC,CAAC;QACjC,CAAC;QAAC,OAAO,GAAQ,EAAE,CAAC;YAClB,IAAI,GAAG,CAAC,IAAI,KAAK,4BAA4B,IAAI,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC;gBACzE,MAAM,IAAI,KAAK,CACb,kCAAkC,QAAQ,IAAI;oBAC9C,gDAAgD;oBAChD,qDAAqD,CACtD,CAAC;YACJ,CAAC;YACD,MAAM,IAAI,KAAK,CAAC,4BAA4B,QAAQ,KAAK,GAAG,CAAC,OAAO,EAAE,CAAC,CAAC;QAC1E,CAAC;IACH,CAAC;IAED,qEAAqE;IACrE,wEAAwE;IACxE,qEAAqE;IACrE,iCAAiC;IACjC,MAAM,SAAS,GAAG,kBAAkB,CAAC,OAAO,CAAC,CAAC;IAE9C,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC3B,MAAM,IAAI,KAAK,CACb,UAAU,QAAQ,qFAAqF,CACxG,CAAC;IACJ,CAAC;IAED,uDAAuD;IACvD,iEAAiE;IACjE,wEAAwE;IACxE,+DAA+D;IAC/D,MAAM,MAAM,GAAqB,SAAS,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;QACpD,GAAG,EAAE;QACL,IAAI,EAAE,mBAAmB,CAAC,EAAE,EAAE,UAAU,CAAC;KAC1C,CAAC,CAAC,CAAC;IAEJ,gEAAgE;IAChE,qEAAqE;IACrE,6DAA6D;IAC7D,MAAM,UAAU,GAAG,IAAI,GAAG,EAAoB,CAAC;IAC/C,KAAK,MAAM,EAAE,IAAI,MAAM,EAAE,CAAC;QACxB,IAAI,EAAE,CAAC,aAAa,EAAE,CAAC;YACrB,MAAM,IAAI,GAAG,UAAU,CAAC,GAAG,CAAC,EAAE,CAAC,aAAa,CAAC,IAAI,EAAE,CAAC;YACpD,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,CAAC;YACnB,UAAU,CAAC,GAAG,CAAC,EAAE,CAAC,aAAa,EAAE,IAAI,CAAC,CAAC;QACzC,CAAC;IACH,CAAC;IAED,wDAAwD;IACxD,yEAAyE;IACzE,4DAA4D;IAC5D,MAAM,KAAK,GAAG,kBAAkB,CAAC,OAAO,CAAC,CAAC;IAE1C,OAAO,EAAE,SAAS,EAAE,MAAM,EAAE,QAAQ,EAAE,OAAO,EAAE,UAAU,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC;AACjF,CAAC"}

package/lib/dist/lib/testCases/types.d.ts

@@ -0,0 +1,242 @@
+import type { TrajectoryStep } from '../../types/index.js';
+import type { TracesAccessor } from '../matchers/index.js';
+/**
+ * Options for a code-based test case.
+ *
+ * All fields are optional. Only the test name (the first argument to
+ * `test()`) is required. When `prompt` is omitted, the runner skips agent
+ * invocation entirely and the test body runs against an empty EvalResult —
+ * useful for purely deterministic / data-driven tests.
+ *
+ * Categories and difficulty levels live in `labels` as prefixed strings
+ * (e.g. `'category:RCA'`, `'difficulty:Medium'`). The legacy top-level
+ * `category` / `difficulty` keys were removed in favor of this single
+ * unified tagging system.
+ */
+export interface TestOptions {
+    /**
+     * Initial prompt sent to the agent. When omitted, the runner does not
+     * invoke the agent and the test body receives an empty EvalResult.
+     */
+    prompt?: string;
+    /** Free-form description shown in the UI. */
+    description?: string;
+    /** Additional context items passed to the agent alongside the prompt. */
+    context?: {
+        description: string;
+        value: string;
+    }[];
+    /**
+     * Labels for filtering and grouping. Use prefixed strings for what was
+     * previously `category` and `difficulty`:
+     * `['category:RCA', 'difficulty:Medium', 'team:platform']`.
+     */
+    labels?: string[];
+    /** Per-test timeout override in milliseconds. */
+    timeout?: number;
+    /**
+     * Plain-text descriptions of expected agent behaviour, used by the
+     * server-side LLM judge and by importers/exporters that round-trip
+     * test cases through JSON. Forwarded to the persisted test case so
+     * server evaluators (`-e <evaluator>`) can grade against the same
+     * criteria a human reviewer would write into the JSON form.
+     *
+     * Inline `judge(result, criteria)` from a test body does NOT need this
+     * field — the body passes the claim directly to /api/judge. Set this
+     * when you want the criteria to (a) live with the test definition for
+     * non-code consumers, or (b) be available to a server-side evaluator
+     * that runs alongside or instead of the body.
+     */
+    expectedOutcomes?: string[];
+    /**
+     * Optional reference trajectory for trajectory-alignment evaluators and
+     * Golden Path comparison. Leave undefined when the test only cares about
+     * outcome quality, not the specific path the agent took.
+     *
+     * Shape mirrors `TestCase.expectedTrajectory` in `@/types` so the
+     * upsert path can forward without coercion.
+     */
+    expectedTrajectory?: {
+        step: number;
+        description: string;
+        requiredTools: string[];
+    }[];
+}
+/**
+ * The trajectory accessor namespace exposed via EvalResult.trajectory in
+ * the new SDK. It IS the same array returned by the runner so users can
+ * iterate freely; the helpers are added as non-enumerable methods so
+ * `for/of` and JSON.stringify behave naturally.
+ */
+export interface TrajectoryAccessor extends Array<TrajectoryStep> {
+    /** Steps of the given type, in order of occurrence. */
+    stepsOfType(type: string): TrajectoryStep[];
+    /** All `action` steps, optionally filtered by toolName and partial args. */
+    toolCalls(name?: string, argsPartial?: Record<string, unknown>): TrajectoryStep[];
+    /** First action-step matching, with `.index` annotated for ordering checks, or null. */
+    firstToolCall(name?: string, argsPartial?: Record<string, unknown>): (TrajectoryStep & {
+        index: number;
+    }) | null;
+}
+export interface CodeTestCase {
+    name: string;
+    options: TestOptions;
+    evaluate: TestBodyFn | LegacyEvaluateFn;
+    /** Resolved file the test was registered from — set by the loader. */
+    sourceFile?: string;
+    /**
+     * Benchmark group path — the joined `describe('...', ...)` names that
+     * wrapped this `test()` call, joined with ' > ' for nested describes.
+     * `undefined` when the test was registered outside any describe; the
+     * loader/CLI then puts the test into the file-default benchmark.
+     */
+    benchmarkPath?: string;
+}
+/**
+ * The Playwright-style test body signature: receives a fixtures object
+ * with `result`, `judge`, `traces`, and `expect`.
+ */
+export type TestBodyFn = (fixtures: TestFixtures) => Promise<void> | void;
+/**
+ * Legacy single-arg signature kept for backward compatibility. Old code
+ * that did `test(name, opts, async (result) => { ... })` keeps working.
+ */
+export type LegacyEvaluateFn = (result: EvalResult) => Promise<void> | void;
+export interface TestFixtures {
+    result: EvalResult;
+    judge: typeof import('./judge.js').judge;
+    traces: TracesAccessor;
+    expect: typeof import('../matchers/expect.js').expect;
+    /**
+     * The `agent` fixture (RFC 004 control inversion). The test body calls
+     * `await agent.run(prompt)` to drive the agent itself, instead of the
+     * framework invoking eagerly before the body. Optional during the
+     * transition: present when the runner provides it; `result` remains for
+     * the legacy eager path. Enforces one invocation per test.
+     */
+    agent?: import('./agentFixture.js').AgentFixture;
+    /**
+     * The `evaluate` fixture (RFC 004 §4.4): run a custom programmatic
+     * evaluator registered via `defineEvaluator()`. Gates by default;
+     * `.observe()` feeds score/insights only.
+     */
+    evaluate?: import('./evaluators.js').EvaluateFn;
+    /**
+     * Static, read-only metadata about the currently-executing test. Useful
+     * for naming temp dirs, logging, and routing — never mutate.
+     */
+    testInfo: TestInfo;
+    /**
+     * Values made available by `beforeEach` hooks via `provide(key, value)`.
+     * Always present (empty when no hooks ran). Read-only from the test body.
+     */
+    provisioned: Readonly<Record<string, unknown>>;
+    /**
+     * Only present inside `beforeEach` hooks. Stores a value into the
+     * provisioned bag for this test. The orchestrator handed to the test
+     * body and `afterEach` does NOT include `provide` (it's `undefined`),
+     * so referencing `fixtures.provide` outside `beforeEach` is a runtime
+     * error — not a no-op. Read provisioned values via
+     * `fixtures.provisioned[key]` instead.
+     */
+    provide?: (key: string, value: unknown) => void;
+}
+/**
+ * Static metadata about a test execution, exposed via `fixtures.testInfo`
+ * and `hookContext.testInfo`. Frozen at the start of the test so hooks and
+ * body see the same values.
+ */
+export interface TestInfo {
+    /** Test name as registered with `test(name, ...)`. */
+    name: string;
+    /**
+     * Joined describe path, e.g. `'A > B'`. `undefined` when the test was
+     * registered outside any describe block.
+     */
+    benchmarkPath?: string;
+    /**
+     * The eval source file the test was loaded from (the absolute path the
+     * loader resolved). `undefined` for tests registered outside the loader
+     * (e.g. unit tests that call `test()` directly).
+     */
+    sourceFile?: string;
+    /**
+     * Storage id of the test case for this run. `undefined` when the
+     * orchestrator runs without a backing storage layer (unit tests).
+     */
+    testCaseId?: string;
+}
+/**
+ * Lifecycle hook kinds, modelled after Playwright. `beforeAll`/`afterAll`
+ * run once per scope (file or describe); `beforeEach`/`afterEach` run
+ * around every matching test in the scope.
+ */
+export type HookKind = 'beforeAll' | 'afterAll' | 'beforeEach' | 'afterEach';
+/**
+ * The function signature accepted by every hook. Receives the same
+ * fixtures object the test body sees. `beforeEach` may call
+ * `fixtures.provide(key, value)` to expose values to the body / `afterEach`.
+ */
+export type HookFn = (fixtures: TestFixtures) => Promise<void> | void;
+/**
+ * One registered hook, captured at registration time. The scope chain is
+ * the joined describe path (`undefined` at file top level), exactly
+ * matching how `test()` records `benchmarkPath`.
+ */
+export interface RegisteredHook {
+    kind: HookKind;
+    fn: HookFn;
+    /** File path the hook was registered from (set by the loader). */
+    sourceFile?: string;
+    /**
+     * Joined describe path the hook lives in, e.g. `'A > B'`. `undefined`
+     * means the hook is file-scoped (outside any describe).
+     */
+    describePath?: string;
+}
+export interface EvalResult {
+    /** All trajectory steps with sugar accessors (toolCalls, firstToolCall, etc.). */
+    trajectory: TrajectoryAccessor;
+    /** Concatenated final response text from the agent's `assistant`/`response` steps. */
+    agentOutput: string;
+    /** Convenience: same as agentOutput. Returns the last assistant text. */
+    finalResponse(): string;
+    /** Try-parse `agentOutput` as JSON. Returns undefined when not parseable. */
+    parsedOutput(): unknown;
+    /** Raw AG-UI events as received from the agent. */
+    rawEvents: any[];
+    /** Agent-supplied run id (for log/trace correlation). */
+    runId?: string;
+    /** Wall-clock duration of the agent invocation in ms (0 when no prompt). */
+    durationMs: number;
+    /** Token usage when reported by the agent. */
+    tokenUsage?: {
+        prompt: number;
+        completion: number;
+        total: number;
+    };
+    /**
+     * OTel traces for this run, available after `agent.run()` resolves
+     * (RFC 004 §4.6). Mirrors the standalone `traces` fixture but scoped to
+     * the result, so `result.traces.totalTokens` reads the same data. Present
+     * only on results returned by `agent.run()`; reading it on the empty
+     * placeholder (before `agent.run()`) yields the loud-failure accessor.
+     */
+    traces?: import('../matchers/traces.js').TracesAccessor;
+    /**
+     * Strategy-C trace-correlation hints (`{ serviceName, startedAt, endedAt }[]`)
+     * the runner computed for this run's agent. Forwarded by `judge()` as
+     * `JudgeRequest.agents` so the agent (trace) judge can find the run's spans
+     * in OpenSearch by service-name + window — the SAME hints the classic
+     * `waitForTracesAndJudge` path builds via `buildJudgeAgentsHints`. This is
+     * what brings SDK `judge()` to parity with the UI/runner trace-judge path
+     * (without it the SDK judge only sends `runId`, which misses subprocess
+     * agents' spans). Present only when the runner could derive a service name.
+     */
+    judgeAgents?: Array<{
+        serviceName: string;
+        startedAt: number;
+        endedAt: number;
+    }>;
+}
+//# sourceMappingURL=types.d.ts.map

package/lib/dist/lib/testCases/types.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../../testCases/types.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,SAAS,CAAC;AAC9C,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AAE3D;;;;;;;;;;;;GAYG;AACH,MAAM,WAAW,WAAW;IAC1B;;;OAGG;IACH,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,6CAA6C;IAC7C,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,yEAAyE;IACzE,OAAO,CAAC,EAAE;QAAE,WAAW,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,EAAE,CAAC;IACnD;;;;OAIG;IACH,MAAM,CAAC,EAAE,MAAM,EAAE,CAAC;IAClB,iDAAiD;IACjD,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB;;;;;;;;;;;;OAYG;IACH,gBAAgB,CAAC,EAAE,MAAM,EAAE,CAAC;IAC5B;;;;;;;OAOG;IACH,kBAAkB,CAAC,EAAE;QACnB,IAAI,EAAE,MAAM,CAAC;QACb,WAAW,EAAE,MAAM,CAAC;QACpB,aAAa,EAAE,MAAM,EAAE,CAAC;KACzB,EAAE,CAAC;CACL;AAED;;;;;GAKG;AACH,MAAM,WAAW,kBAAmB,SAAQ,KAAK,CAAC,cAAc,CAAC;IAC/D,uDAAuD;IACvD,WAAW,CAAC,IAAI,EAAE,MAAM,GAAG,cAAc,EAAE,CAAC;IAC5C,4EAA4E;IAC5E,SAAS,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,WAAW,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,cAAc,EAAE,CAAC;IAClF,wFAAwF;IACxF,aAAa,CACX,IAAI,CAAC,EAAE,MAAM,EACb,WAAW,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GACpC,CAAC,cAAc,GAAG;QAAE,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC,GAAG,IAAI,CAAC;CAChD;AAED,MAAM,WAAW,YAAY;IAC3B,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,WAAW,CAAC;IACrB,QAAQ,EAAE,UAAU,GAAG,gBAAgB,CAAC;IACxC,sEAAsE;IACtE,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB;;;;;OAKG;IACH,aAAa,CAAC,EAAE,MAAM,CAAC;CACxB;AAED;;;GAGG;AACH,MAAM,MAAM,UAAU,GAAG,CAAC,QAAQ,EAAE,YAAY,KAAK,OAAO,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC;AAE1E;;;GAGG;AACH,MAAM,MAAM,gBAAgB,GAAG,CAAC,MAAM,EAAE,UAAU,KAAK,OAAO,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC;AAE5E,MAAM,WAAW,YAAY;IAC3B,MAAM,EAAE,UAAU,CAAC;IACnB,KAAK,EAAE,cAAc,YAAY,EAAE,KAAK,CAAC;IACzC,MAAM,EAAE,cAAc,CAAC;IACvB,MAAM,EAAE,cAAc,uBAAuB,EAAE,MAAM,CAAC;IACtD;;;;;;OAMG;IACH,KAAK,CAAC,EAAE,OAAO,mBAAmB,EAAE,YAAY,CAAC;IACjD;;;;OAIG;IACH,QAAQ,CAAC,EAAE,OAAO,iBAAiB,EAAE,UAAU,CAAC;IAChD;;;OAGG;IACH,QAAQ,EAAE,QAAQ,CAAC;IACnB;;;OAGG;IACH,WAAW,EAAE,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC;IAC/C;;;;;;;OAOG;IACH,OAAO,CAAC,EAAE,CAAC,GAAG,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,KAAK,IAAI,CAAC;CACjD;AAED;;;;GAIG;AACH,MAAM,WAAW,QAAQ;IACvB,sDAAsD;IACtD,IAAI,EAAE,MAAM,CAAC;IACb;;;OAGG;IACH,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB;;;;OAIG;IACH,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB;;;OAGG;IACH,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED;;;;GAIG;AACH,MAAM,MAAM,QAAQ,GAAG,WAAW,GAAG,UAAU,GAAG,YAAY,GAAG,WAAW,CAAC;AAE7E;;;;GAIG;AACH,MAAM,MAAM,MAAM,GAAG,CAAC,QAAQ,EAAE,YAAY,KAAK,OAAO,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC;AAEtE;;;;GAIG;AACH,MAAM,WAAW,cAAc;IAC7B,IAAI,EAAE,QAAQ,CAAC;IACf,EAAE,EAAE,MAAM,CAAC;IACX,kEAAkE;IAClE,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB;;;OAGG;IACH,YAAY,CAAC,EAAE,MAAM,CAAC;CACvB;AAED,MAAM,WAAW,UAAU;IACzB,kFAAkF;IAClF,UAAU,EAAE,kBAAkB,CAAC;IAC/B,sFAAsF;IACtF,WAAW,EAAE,MAAM,CAAC;IACpB,yEAAyE;IACzE,aAAa,IAAI,MAAM,CAAC;IACxB,6EAA6E;IAC7E,YAAY,IAAI,OAAO,CAAC;IACxB,mDAAmD;IACnD,SAAS,EAAE,GAAG,EAAE,CAAC;IACjB,yDAAyD;IACzD,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,4EAA4E;IAC5E,UAAU,EAAE,MAAM,CAAC;IACnB,8CAA8C;IAC9C,UAAU,CAAC,EAAE;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,UAAU,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC;IACnE;;;;;;OAMG;IACH,MAAM,CAAC,EAAE,OAAO,uBAAuB,EAAE,cAAc,CAAC;IAExD;;;;;;;;;OASG;IACH,WAAW,CAAC,EAAE,KAAK,CAAC;QAAE,WAAW,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;CAClF"}

package/lib/dist/lib/testCases/types.js

@@ -0,0 +1,6 @@
+/*
+ * Copyright OpenSearch Contributors
+ * SPDX-License-Identifier: Apache-2.0
+ */
+export {};
+//# sourceMappingURL=types.js.map

package/lib/dist/lib/testCases/types.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"types.js","sourceRoot":"","sources":["../../../testCases/types.ts"],"names":[],"mappings":"AAAA;;;GAGG"}

package/lib/dist/lib/theme.d.ts

@@ -0,0 +1,6 @@
+export type Theme = 'light' | 'dark';
+export declare function getTheme(): Theme;
+export declare function setTheme(theme: Theme): void;
+export declare function applyTheme(theme: Theme): void;
+export declare function initializeTheme(): void;
+//# sourceMappingURL=theme.d.ts.map

package/lib/dist/lib/theme.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"theme.d.ts","sourceRoot":"","sources":["../../theme.ts"],"names":[],"mappings":"AAKA,MAAM,MAAM,KAAK,GAAG,OAAO,GAAG,MAAM,CAAC;AAIrC,wBAAgB,QAAQ,IAAI,KAAK,CAShC;AAED,wBAAgB,QAAQ,CAAC,KAAK,EAAE,KAAK,GAAG,IAAI,CAK3C;AAED,wBAAgB,UAAU,CAAC,KAAK,EAAE,KAAK,GAAG,IAAI,CAU7C;AAED,wBAAgB,eAAe,IAAI,IAAI,CAGtC"}

package/lib/dist/lib/theme.js

@@ -0,0 +1,36 @@
+/*
+ * Copyright OpenSearch Contributors
+ * SPDX-License-Identifier: Apache-2.0
+ */
+const THEME_STORAGE_KEY = 'agent-health-theme';
+export function getTheme() {
+    if (typeof window === 'undefined')
+        return 'dark';
+    const stored = localStorage.getItem(THEME_STORAGE_KEY);
+    if (stored === 'light' || stored === 'dark') {
+        return stored;
+    }
+    return 'dark';
+}
+export function setTheme(theme) {
+    if (typeof window === 'undefined')
+        return;
+    localStorage.setItem(THEME_STORAGE_KEY, theme);
+    applyTheme(theme);
+}
+export function applyTheme(theme) {
+    if (typeof window === 'undefined')
+        return;
+    const root = document.documentElement;
+    if (theme === 'dark') {
+        root.classList.add('dark');
+    }
+    else {
+        root.classList.remove('dark');
+    }
+}
+export function initializeTheme() {
+    const theme = getTheme();
+    applyTheme(theme);
+}
+//# sourceMappingURL=theme.js.map

package/lib/dist/lib/theme.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"theme.js","sourceRoot":"","sources":["../../theme.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAIH,MAAM,iBAAiB,GAAG,oBAAoB,CAAC;AAE/C,MAAM,UAAU,QAAQ;IACtB,IAAI,OAAO,MAAM,KAAK,WAAW;QAAE,OAAO,MAAM,CAAC;IAEjD,MAAM,MAAM,GAAG,YAAY,CAAC,OAAO,CAAC,iBAAiB,CAAC,CAAC;IACvD,IAAI,MAAM,KAAK,OAAO,IAAI,MAAM,KAAK,MAAM,EAAE,CAAC;QAC5C,OAAO,MAAM,CAAC;IAChB,CAAC;IAED,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,MAAM,UAAU,QAAQ,CAAC,KAAY;IACnC,IAAI,OAAO,MAAM,KAAK,WAAW;QAAE,OAAO;IAE1C,YAAY,CAAC,OAAO,CAAC,iBAAiB,EAAE,KAAK,CAAC,CAAC;IAC/C,UAAU,CAAC,KAAK,CAAC,CAAC;AACpB,CAAC;AAED,MAAM,UAAU,UAAU,CAAC,KAAY;IACrC,IAAI,OAAO,MAAM,KAAK,WAAW;QAAE,OAAO;IAE1C,MAAM,IAAI,GAAG,QAAQ,CAAC,eAAe,CAAC;IAEtC,IAAI,KAAK,KAAK,MAAM,EAAE,CAAC;QACrB,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;IAC7B,CAAC;SAAM,CAAC;QACN,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;IAChC,CAAC;AACH,CAAC;AAED,MAAM,UAAU,eAAe;IAC7B,MAAM,KAAK,GAAG,QAAQ,EAAE,CAAC;IACzB,UAAU,CAAC,KAAK,CAAC,CAAC;AACpB,CAAC"}

package/lib/dist/lib/uiTelemetry.d.ts

@@ -0,0 +1,7 @@
+/**
+ * Fire-and-forget UI usage telemetry. Lets us understand load/usage patterns
+ * (e.g. which comparison search scope is used, how often, result counts)
+ * without a heavyweight analytics dependency. Never blocks the UI or throws.
+ */
+export declare function recordUiEvent(event: string, props?: Record<string, unknown>): void;
+//# sourceMappingURL=uiTelemetry.d.ts.map

package/lib/dist/lib/uiTelemetry.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"uiTelemetry.d.ts","sourceRoot":"","sources":["../../uiTelemetry.ts"],"names":[],"mappings":"AAKA;;;;GAIG;AACH,wBAAgB,aAAa,CAAC,KAAK,EAAE,MAAM,EAAE,KAAK,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,IAAI,CAalF"}

package/lib/dist/lib/uiTelemetry.js

@@ -0,0 +1,25 @@
+/*
+ * Copyright OpenSearch Contributors
+ * SPDX-License-Identifier: Apache-2.0
+ */
+/**
+ * Fire-and-forget UI usage telemetry. Lets us understand load/usage patterns
+ * (e.g. which comparison search scope is used, how often, result counts)
+ * without a heavyweight analytics dependency. Never blocks the UI or throws.
+ */
+export function recordUiEvent(event, props) {
+    try {
+        void fetch('/api/telemetry/ui-event', {
+            method: 'POST',
+            headers: { 'Content-Type': 'application/json' },
+            body: JSON.stringify({ event, props }),
+            keepalive: true,
+        }).catch(() => {
+            /* telemetry is best-effort */
+        });
+    }
+    catch {
+        /* never let telemetry break the UI */
+    }
+}
+//# sourceMappingURL=uiTelemetry.js.map

package/lib/dist/lib/uiTelemetry.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"uiTelemetry.js","sourceRoot":"","sources":["../../uiTelemetry.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH;;;;GAIG;AACH,MAAM,UAAU,aAAa,CAAC,KAAa,EAAE,KAA+B;IAC1E,IAAI,CAAC;QACH,KAAK,KAAK,CAAC,yBAAyB,EAAE;YACpC,MAAM,EAAE,MAAM;YACd,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE;YAC/C,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,KAAK,EAAE,CAAC;YACtC,SAAS,EAAE,IAAI;SAChB,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE;YACZ,8BAA8B;QAChC,CAAC,CAAC,CAAC;IACL,CAAC;IAAC,MAAM,CAAC;QACP,sCAAsC;IACxC,CAAC;AACH,CAAC"}

package/lib/dist/lib/utils.d.ts

@@ -0,0 +1,111 @@
+import { type ClassValue } from "clsx";
+import { Difficulty, DateFormatVariant } from "../types/index.js";
+export declare function cn(...inputs: ClassValue[]): string;
+/**
+ * Returns Tailwind classes for styling label badges
+ * Only difficulty:Easy/Medium/Hard get special colors; all others use hash-based palette
+ */
+export declare const getLabelColor: (label: string) => string;
+/**
+ * Returns Tailwind classes for styling difficulty badges
+ * @deprecated Use getLabelColor with difficulty: prefixed labels instead
+ */
+export declare const getDifficultyColor: (difficulty: Difficulty) => string;
+/**
+ * Formats a timestamp string to a localized date string
+ * @param timestamp - ISO timestamp string
+ * @param variant - 'date' (date only), 'datetime' (default, with time), 'detailed' (with seconds)
+ */
+export declare const formatDate: (timestamp: string, variant?: DateFormatVariant) => string;
+/**
+ * Formats a timestamp to relative time (e.g., "5m ago", "2h ago")
+ * Falls back to formatDate for timestamps older than 7 days
+ */
+export declare const formatRelativeTime: (timestamp: string) => string;
+/**
+ * Gets the display name for a model ID from config
+ */
+export declare const getModelName: (modelId: string) => string;
+/**
+ * Returns a short, stable suffix for a TestCaseRun id, suitable for embedding
+ * in an auto-generated run name. Run ids are shaped
+ * `report-<timestamp>-<random>` (file adapter) or arbitrary keys (OpenSearch),
+ * so we take the trailing 6 characters which gives a recognizable yet compact
+ * label without exposing the full id. Falls back to the full id for very short
+ * ids (e.g. legacy data).
+ */
+export declare const getRunShortId: (runId: string) => string;
+/**
+ * Resolves the display name for a TestCaseRun.
+ *
+ * Prefers the persisted `name` field (set from the run config dialog or
+ * auto-generated server-side). Falls back to `Run <short-id>` for older
+ * runs that pre-date the field, so every row in the runs list has a
+ * recognizable label instead of a raw id slice.
+ */
+export declare const getRunDisplayName: (run: {
+    id: string;
+    name?: string;
+}) => string;
+/**
+ * Computes a single "overall score" percentage to display next to a run in
+ * a list, from whatever metrics that run happens to carry.
+ *
+ * Why this is non-trivial:
+ *   - Only the *RCA Default* system evaluator emits a metric named `accuracy`.
+ *     Other system evaluators (Factuality, Tool Use, Reasoning, Safety) and
+ *     any custom evaluator emit completely different metric names like
+ *     `tool_selection_accuracy`, `reasoning_coherence`, `bias_detection`, etc.
+ *   - The runs list used to hardcode `run.metrics?.accuracy ?? 0`, which made
+ *     every non-RCA run display `0%` even when the judge had passed it.
+ *
+ * The fix is to take the **arithmetic mean of all populated numeric metrics**
+ * on the run. This works for any evaluator without having to look up its
+ * `scoringConfig`, gives a reasonable summary number, and degrades to `null`
+ * (rendered as `—`) when no metrics are present — the only honest answer
+ * for runs whose judge call hasn't completed or which were never scored.
+ *
+ * Returns `null` when there are no numeric metrics; otherwise an integer
+ * percentage rounded to the nearest whole number.
+ */
+export declare const getRunOverallScore: (metrics: Record<string, number | undefined> | undefined | null) => number | null;
+/**
+ * Returns theme-aware Tailwind classes for pass rate badges
+ * Light mode: More saturated backgrounds with darker text for better visibility
+ * Dark mode: Lighter text on dark backgrounds
+ */
+export declare const getPassRateColor: (passRate: number) => string;
+/**
+ * Returns theme-aware Tailwind classes for status indicators
+ * Used for success/error/warning/info states
+ * Light mode: More saturated backgrounds (100 shade) for better visibility
+ * Dark mode: Semi-transparent darker backgrounds with lighter text
+ */
+export declare const getStatusColor: (status: "success" | "error" | "warning" | "info" | "neutral") => string;
+/**
+ * Returns theme-aware text color classes for metrics
+ * Used for displaying metric values in stats cards
+ * Light mode: Darker colors for WCAG AA compliance (4.5:1 contrast ratio)
+ * Dark mode: Lighter colors for visibility on dark backgrounds
+ */
+export declare const getMetricTextColor: (type: "primary" | "success" | "error" | "warning" | "info" | "secondary") => string;
+/**
+ * Truncates text to a specified length with ellipsis
+ */
+export declare const truncate: (text: string, length: number) => string;
+/**
+ * Detect the syntax-highlighting language for a code-SDK eval file from its
+ * extension. Isomorphic (no Node built-ins) so it's shared by the CLI/server
+ * import path (lib/testCases/loader.ts, which re-exports this) AND the
+ * browser-side EvalSourceCodeView component -- one source of truth for
+ * "what language is this file" instead of duplicating the extension check.
+ *
+ * `.mjs`/`.js`/`.cjs` -> javascript, everything else code-like (`.ts` and
+ * unknown extensions) -> typescript. There's no `.jsx`/`.tsx` case today --
+ * eval files are plain Node scripts, not React -- but typescript's grammar
+ * is a superset of JS syntax so defaulting unknown-but-code extensions to
+ * it is the safer guess for highlighting purposes (worst case: a few
+ * JS-only tokens render unstyled, never mis-highlighted).
+ */
+export declare function detectSourceLanguage(fileName: string): 'javascript' | 'typescript';
+//# sourceMappingURL=utils.d.ts.map