@forwardimpact/libeval 0.1.0 → 0.1.1

package/src/tee-writer.js

@@ -0,0 +1,157 @@
+/**
+ * TeeWriter — a Writable stream that writes raw NDJSON to a file while
+ * simultaneously streaming human-readable text to a separate stream (e.g.
+ * process.stdout).
+ *
+ * Supports two modes:
+ * - "raw" (default): expects standard stream-json events from AgentRunner
+ * - "supervised": expects tagged events {source, turn, event} from Supervisor
+ *
+ * Follows OO+DI: constructor injection, factory function, tests bypass factory.
+ */
+
+import { Writable } from "node:stream";
+import { TraceCollector } from "./trace-collector.js";
+
+export class TeeWriter extends Writable {
+  /**
+   * @param {object} deps
+   * @param {import("stream").Writable} deps.fileStream - Stream to write raw NDJSON to
+   * @param {import("stream").Writable} deps.textStream - Stream to write human-readable text to
+   * @param {"raw"|"supervised"} [deps.mode] - Event format: "raw" or "supervised" (default: "raw")
+   */
+  constructor({ fileStream, textStream, mode }) {
+    super();
+    if (!fileStream) throw new Error("fileStream is required");
+    if (!textStream) throw new Error("textStream is required");
+    this.fileStream = fileStream;
+    this.textStream = textStream;
+    this.mode = mode ?? "raw";
+    this.collector = new TraceCollector();
+    this.turnsEmitted = 0;
+    this.lastSource = null;
+    this.partial = "";
+  }
+
+  /**
+   * @param {Buffer|string} chunk
+   * @param {string} encoding
+   * @param {function} callback
+   */
+  _write(chunk, encoding, callback) {
+    const str = this.partial + chunk.toString();
+    const lines = str.split("\n");
+    this.partial = lines.pop() ?? "";
+
+    for (const line of lines) {
+      if (!line.trim()) continue;
+      this.fileStream.write(line + "\n");
+      this.processLine(line);
+    }
+    callback();
+  }
+
+  /**
+   * @param {function} callback
+   */
+  _final(callback) {
+    if (this.partial.trim()) {
+      this.fileStream.write(this.partial + "\n");
+      this.processLine(this.partial);
+    }
+
+    if (this.mode === "raw" && this.collector.result) {
+      const text = this.collector.toText();
+      const idx = text.lastIndexOf("\n---");
+      if (idx !== -1) {
+        this.textStream.write(text.slice(idx) + "\n");
+      }
+    }
+
+    callback();
+  }
+
+  /**
+   * Process a single NDJSON line — feed to collector and flush text.
+   * @param {string} line
+   */
+  processLine(line) {
+    if (this.mode === "supervised") {
+      this.processSupervisedLine(line);
+    } else {
+      this.collector.addLine(line);
+      this.flushTurns();
+    }
+  }
+
+  /**
+   * Handle a tagged supervisor line: unwrap event, show source labels.
+   * @param {string} line
+   */
+  processSupervisedLine(line) {
+    let parsed;
+    try {
+      parsed = JSON.parse(line);
+    } catch {
+      return;
+    }
+
+    if (parsed.source === "orchestrator" && parsed.type === "summary") {
+      const status = parsed.success ? "completed" : "incomplete";
+      this.textStream.write(
+        `\n--- Evaluation ${status} after ${parsed.turns} turns ---\n`,
+      );
+      return;
+    }
+
+    if (parsed.event) {
+      if (parsed.source && parsed.source !== this.lastSource) {
+        this.lastSource = parsed.source;
+        this.textStream.write(`\n[${parsed.source}]\n`);
+      }
+      this.collector.addLine(JSON.stringify(parsed.event));
+      this.flushTurns();
+    }
+  }
+
+  /**
+   * Emit text for any new turns accumulated by the collector.
+   */
+  flushTurns() {
+    const turns = this.collector.turns;
+    while (this.turnsEmitted < turns.length) {
+      const turn = turns[this.turnsEmitted++];
+      if (turn.role === "assistant") {
+        for (const block of turn.content) {
+          if (block.type === "text") {
+            this.textStream.write(block.text + "\n");
+          } else if (block.type === "tool_use") {
+            const input = summarizeInput(block.input);
+            this.textStream.write(`> Tool: ${block.name} ${input}\n`);
+          }
+        }
+      }
+    }
+  }
+}
+
+/**
+ * Summarize tool input for text display, truncated to keep logs readable.
+ * @param {object} input - Tool input object
+ * @returns {string} Truncated summary
+ */
+function summarizeInput(input) {
+  if (!input || typeof input !== "object") return "";
+  const json = JSON.stringify(input);
+  if (json.length <= 200) return json;
+  return json.slice(0, 197) + "...";
+}
+
+/**
+ * Factory function — wires a TeeWriter with the given streams.
+ * @param {object} deps - Same as TeeWriter constructor
+ * @returns {TeeWriter}
+ */
+export function createTeeWriter(deps) {
+  return new TeeWriter(deps);
+}

package/test/agent-runner.test.js

@@ -0,0 +1,292 @@
+import { describe, test } from "node:test";
+import assert from "node:assert";
+import { PassThrough } from "node:stream";
+
+import { AgentRunner, createAgentRunner } from "@forwardimpact/libeval";
+
+/**
+ * Create a mock query function that yields canned messages.
+ * @param {object[]} messages - Messages to yield
+ * @param {function} [captureOptions] - Callback to capture query options
+ * @returns {function}
+ */
+function mockQuery(messages, captureOptions) {
+  return async function* (params) {
+    if (captureOptions) captureOptions(params);
+    for (const msg of messages) {
+      yield msg;
+    }
+  };
+}
+
+/**
+ * Collect all NDJSON lines written to a PassThrough stream.
+ * @param {PassThrough} stream
+ * @returns {string[]}
+ */
+function collectLines(stream) {
+  const data = stream.read();
+  if (!data) return [];
+  return data
+    .toString()
+    .trim()
+    .split("\n")
+    .filter((l) => l.length > 0);
+}
+
+describe("AgentRunner", () => {
+  test("constructor throws on missing cwd", () => {
+    assert.throws(
+      () =>
+        new AgentRunner({
+          query: async function* () {},
+          output: new PassThrough(),
+        }),
+      /cwd is required/,
+    );
+  });
+
+  test("constructor throws on missing query", () => {
+    assert.throws(
+      () => new AgentRunner({ cwd: "/tmp", output: new PassThrough() }),
+      /query is required/,
+    );
+  });
+
+  test("constructor throws on missing output", () => {
+    assert.throws(
+      () =>
+        new AgentRunner({
+          cwd: "/tmp",
+          query: async function* () {},
+        }),
+      /output is required/,
+    );
+  });
+
+  test("constructor uses defaults for optional params", () => {
+    const runner = new AgentRunner({
+      cwd: "/tmp",
+      query: async function* () {},
+      output: new PassThrough(),
+    });
+    assert.strictEqual(runner.model, "opus");
+    assert.strictEqual(runner.maxTurns, 50);
+    assert.deepStrictEqual(runner.allowedTools, [
+      "Bash",
+      "Read",
+      "Glob",
+      "Grep",
+      "Write",
+      "Edit",
+    ]);
+    assert.strictEqual(runner.permissionMode, "bypassPermissions");
+    assert.strictEqual(runner.sessionId, null);
+  });
+
+  test("run() writes NDJSON lines to output stream", async () => {
+    const messages = [
+      { type: "system", subtype: "init", session_id: "sess-1" },
+      { type: "assistant", content: "Working on it..." },
+      { type: "result", subtype: "success", result: "Done." },
+    ];
+
+    const output = new PassThrough();
+    const runner = new AgentRunner({
+      cwd: "/tmp",
+      query: mockQuery(messages),
+      output,
+    });
+
+    const result = await runner.run("Test task");
+    const lines = collectLines(output);
+
+    assert.strictEqual(lines.length, 3);
+    assert.deepStrictEqual(JSON.parse(lines[0]), messages[0]);
+    assert.deepStrictEqual(JSON.parse(lines[1]), messages[1]);
+    assert.deepStrictEqual(JSON.parse(lines[2]), messages[2]);
+    assert.strictEqual(result.success, true);
+    assert.strictEqual(result.text, "Done.");
+    assert.strictEqual(result.sessionId, "sess-1");
+  });
+
+  test("run() captures sessionId from init event", async () => {
+    const messages = [
+      { type: "system", subtype: "init", session_id: "my-session" },
+      { type: "result", subtype: "success", result: "OK" },
+    ];
+
+    const output = new PassThrough();
+    const runner = new AgentRunner({
+      cwd: "/tmp",
+      query: mockQuery(messages),
+      output,
+    });
+
+    await runner.run("Task");
+    assert.strictEqual(runner.sessionId, "my-session");
+  });
+
+  test("run() passes options to query", async () => {
+    let captured = null;
+    const query = mockQuery(
+      [{ type: "result", subtype: "success", result: "OK" }],
+      (params) => {
+        captured = params;
+      },
+    );
+
+    const output = new PassThrough();
+    const runner = new AgentRunner({
+      cwd: "/work",
+      query,
+      output,
+      model: "sonnet",
+      maxTurns: 10,
+      allowedTools: ["Read", "Grep"],
+      permissionMode: "plan",
+    });
+
+    await runner.run("My task");
+
+    assert.strictEqual(captured.prompt, "My task");
+    assert.strictEqual(captured.options.cwd, "/work");
+    assert.strictEqual(captured.options.model, "sonnet");
+    assert.strictEqual(captured.options.maxTurns, 10);
+    assert.deepStrictEqual(captured.options.allowedTools, ["Read", "Grep"]);
+    assert.strictEqual(captured.options.permissionMode, "plan");
+    assert.strictEqual(captured.options.allowDangerouslySkipPermissions, true);
+  });
+
+  test("run() returns success=false on non-success subtype", async () => {
+    const messages = [{ type: "result", subtype: "error", result: "Stopped" }];
+
+    const output = new PassThrough();
+    const runner = new AgentRunner({
+      cwd: "/tmp",
+      query: mockQuery(messages),
+      output,
+    });
+
+    const result = await runner.run("Task");
+    assert.strictEqual(result.success, false);
+    assert.strictEqual(result.text, "Stopped");
+  });
+
+  test("resume() passes sessionId via options.resume", async () => {
+    let resumeCapture = null;
+
+    const initMessages = [
+      { type: "system", subtype: "init", session_id: "sess-42" },
+      { type: "result", subtype: "success", result: "First done" },
+    ];
+
+    let callCount = 0;
+    const query = async function* (params) {
+      callCount++;
+      if (callCount === 1) {
+        for (const m of initMessages) yield m;
+      } else {
+        resumeCapture = params;
+        yield { type: "result", subtype: "success", result: "Resumed" };
+      }
+    };
+
+    const output = new PassThrough();
+    const runner = new AgentRunner({ cwd: "/tmp", query, output });
+
+    await runner.run("Initial task");
+    const result = await runner.resume("Follow up");
+
+    assert.strictEqual(resumeCapture.options.resume, "sess-42");
+    assert.strictEqual(resumeCapture.prompt, "Follow up");
+    assert.strictEqual(result.success, true);
+    assert.strictEqual(result.text, "Resumed");
+  });
+
+  test("drainOutput() returns buffered lines and clears buffer", async () => {
+    const messages = [
+      { type: "assistant", content: "Line 1" },
+      { type: "result", subtype: "success", result: "Line 2" },
+    ];
+
+    const output = new PassThrough();
+    const runner = new AgentRunner({
+      cwd: "/tmp",
+      query: mockQuery(messages),
+      output,
+    });
+
+    await runner.run("Task");
+
+    const drained = runner.drainOutput();
+    assert.strictEqual(drained.length, 2);
+    assert.deepStrictEqual(JSON.parse(drained[0]), messages[0]);
+    assert.deepStrictEqual(JSON.parse(drained[1]), messages[1]);
+
+    // Buffer should be empty after drain
+    const secondDrain = runner.drainOutput();
+    assert.strictEqual(secondDrain.length, 0);
+  });
+
+  test("run() captures error when query throws and returns buffered output", async () => {
+    async function* failingQuery() {
+      yield { type: "system", subtype: "init", session_id: "sess-err" };
+      yield { type: "assistant", content: "Partial work" };
+      throw new Error("Claude Code process exited with code 1");
+    }
+
+    const output = new PassThrough();
+    const runner = new AgentRunner({
+      cwd: "/tmp",
+      query: () => failingQuery(),
+      output,
+    });
+
+    const result = await runner.run("Task");
+    assert.strictEqual(result.success, false);
+    assert.ok(result.error);
+    assert.match(result.error.message, /exited with code 1/);
+    assert.strictEqual(result.sessionId, "sess-err");
+
+    // Buffered output should contain the messages yielded before the error
+    const drained = runner.drainOutput();
+    assert.strictEqual(drained.length, 2);
+  });
+
+  test("resume() captures error when query throws", async () => {
+    const initMessages = [
+      { type: "system", subtype: "init", session_id: "sess-r" },
+      { type: "result", subtype: "success", result: "OK" },
+    ];
+
+    let callCount = 0;
+    const query = async function* () {
+      callCount++;
+      if (callCount === 1) {
+        for (const m of initMessages) yield m;
+      } else {
+        yield { type: "assistant", content: "Resuming..." };
+        throw new Error("Process crashed");
+      }
+    };
+
+    const output = new PassThrough();
+    const runner = new AgentRunner({ cwd: "/tmp", query, output });
+
+    await runner.run("Task");
+    const result = await runner.resume("Continue");
+    assert.strictEqual(result.success, false);
+    assert.ok(result.error);
+    assert.match(result.error.message, /Process crashed/);
+  });
+
+  test("createAgentRunner factory returns an AgentRunner instance", () => {
+    const runner = createAgentRunner({
+      cwd: "/tmp",
+      query: async function* () {},
+      output: new PassThrough(),
+    });
+    assert.ok(runner instanceof AgentRunner);
+  });
+});