Build minimal LLM eval harness scaffold (B5) — fixtures recorded separately

Land the B5 harness skeleton: fixture loader (strict JSONL contract with
meta header + ordered turn records + assertions tail), replay-client that
swaps in for pi-ai's streamSimple by replaying recorded assistant turns,
structural assertion judge (tool_called / no_tool_called /
message_contains_substring / tool_call_arg_matches / trace_max_turns), an
eval-runner that sums the token ledger and produces an EvalResult, a
vitest driver (tests/eval/harness.test.ts) that discovers *.jsonl
fixtures, and a record-fixture CLI that writes fixture meta headers.

Hard contracts:
- No fallback: replay-client and fixture-loader throw on exhaustion /
  malformed input rather than synthesizing defaults.
- No mock: types reuse @mariozechner/pi-ai AgentMessage / ToolCall
  verbatim; shadow types are forbidden.
- No fake fixture data: tests/eval/fixtures/ ships only README +
  .gitkeep. Real traces are recorded by the user against live providers.

package.json adds test:eval (vitest --passWithNoTests so an empty
fixture dir is green) and eval:record (CLI entrypoint). electron-builder
already excludes tests/ via its files: glob — fixtures never enter the
shipped .asar.

tests/eval/harness.test.ts

@@ -0,0 +1,49 @@
+/**
+ * B5 eval harness — driver spec.
+ *
+ * This spec discovers every `*.jsonl` fixture under `tests/eval/fixtures/`
+ * and runs it through the assertion judge. If there are zero fixtures, the
+ * spec file has zero tests — vitest is invoked with `--passWithNoTests`
+ * from the `test:eval` script so this is a green no-op until the user
+ * records real fixtures.
+ *
+ * The harness does NOT synthesize placeholder fixtures under any
+ * circumstance. An empty `fixtures/` directory is a real state — the tier
+ * is simply not yet exercising anything — and the green run reflects that
+ * honestly.
+ */
+
+import { readdir } from "node:fs/promises";
+import { join } from "node:path";
+import { describe, expect, it } from "vitest";
+import { loadFixture } from "./fixture-loader.js";
+import { runEval } from "./eval-runner.js";
+
+const FIXTURE_DIR = join(__dirname, "fixtures");
+
+async function listFixtures(): Promise<string[]> {
+  const entries = await readdir(FIXTURE_DIR, { withFileTypes: true });
+  return entries
+    .filter((entry) => entry.isFile() && entry.name.endsWith(".jsonl"))
+    .map((entry) => join(FIXTURE_DIR, entry.name))
+    .sort();
+}
+
+const fixturePaths = await listFixtures();
+
+describe.each(fixturePaths)("eval fixture %s", (fixturePath) => {
+  it("passes all assertions", async () => {
+    const fixture = await loadFixture(fixturePath);
+    const result = runEval(fixture);
+    if (!result.passed) {
+      const failures = result.assertions
+        .filter((assertion) => !assertion.passed)
+        .map((assertion) => `- ${assertion.assertion.kind}: ${assertion.reason ?? "<no reason>"}`)
+        .join("\n");
+      throw new Error(
+        `fixture ${result.fixtureId} failed:\n${failures}\ntoken ledger: ${JSON.stringify(result.tokenLedger)}`,
+      );
+    }
+    expect(result.passed).toBe(true);
+  });
+});