char-slop/berry

git clone https://git.t4t.associates/char-slop/berry

Charlotte Sommove computer-use out of the repof5275c62026-10-11

import { assert, assertEquals, assertStringIncludes } from "@std/assert";
import { BACKGROUND_CONTEXT as context } from "@earendil-works/chord/context";
import { type Message } from "@earendil-works/pi-ai";
import { fauxAssistantMessage, fauxToolCall } from "@earendil-works/pi-ai/providers/faux";
import { defineExtension } from "@berry/api";
import { until } from "@berry/util";
import { lastToolResult, script, setup, toolsOffered, turn } from "./fixture.ts";

const images = (messages: readonly Message[]) =>
  messages.flatMap((message) =>
    message.role === "toolResult"
      ? message.content.flatMap((block) => block.type === "image" ? [block.data] : [])
      : []
  );

Deno.test("selected context hooks run in order on every request, compaction's too, without changing what is stored", async () => {
  await using harness = await setup();
  const { berry, faux, model } = harness;
  for (const name of ["first", "second", "unused", "refuse"]) {
    await berry.install(defineExtension({
      name,
      initial: () => null,
      setup() {
        return (conv) =>
          conv.onContext(async (messages, requestedModel) => {
            await Promise.resolve();
            if (name === "unused") throw new Error("An unselected hook ran");
            if (name === "refuse") throw new Error("context refused");
            assertEquals(requestedModel.id, faux.getModel().id);
            // the second sees what the first returned
            return messages.map((message) =>
              message.role !== "user" ? message : {
                ...message,
                content: name === "first" ? "filtered" : `${message.content} second`,
              }
            );
          });
      },
    }));
  }
  const tid = await berry.create({ extensions: { first: {}, second: {} }, model });
  const conversation = (await berry.conversation(tid))!;
  const longInput = "x".repeat(100_000);
  const [seen] = await turn(harness, tid, []);
  assertEquals(seen.findLast((message) => message.role === "user")?.content, "filtered second");
  assertEquals(toolsOffered(seen), ["codemode"]);

  faux.setResponses([fauxAssistantMessage("done")]);
  await (await conversation.submit({ type: "input", content: longInput }, context)).wait(context);
  await conversation.waitForIdle(context);
  let summarized: readonly Message[] = [];
  faux.setResponses([(request) => {
    summarized = request.messages;
    return fauxAssistantMessage("A summary");
  }]);
  await berry.harness.waitForTask(await conversation.compact(undefined, context), context);
  assertEquals(
    summarized.findLast((message) => message.role === "user")?.content,
    "filtered second",
  );
  const stored = await conversation.context(context);
  assertEquals(
    stored.messages.findLast((message) => message.role === "user")?.content,
    longInput,
  );

  // a failing hook ends the request without calling the provider
  const refused = await berry.create({ extensions: { refuse: {} }, model });
  const calls = faux.state.callCount;
  assertEquals(await turn(harness, refused, []), []);
  assertEquals(faux.state.callCount, calls);
  const entries = (await (await berry.conversation(refused))!.entries(
    { order: "ascending" },
    100,
    undefined,
    context,
  ))
    .items;
  const failure = entries.flatMap((entry) => entry.model ?? []).findLast((message) =>
    message.role === "assistant"
  );
  assert(failure?.role === "assistant");
  assertEquals(failure.stopReason, "error");
  assertStringIncludes(failure.errorMessage ?? "", "context refused");
});

Deno.test("command context appends passively with its event, and busy writes wait for the boundary", async () => {
  await using harness = await setup();
  const { berry, model, faux, gate } = harness;
  await berry.install(defineExtension<string[], string>({
    name: "reports",
    initial: () => [],
    apply: (state, event) => [...state, event],
    setup() {
      return (conv) =>
        conv.command({
          name: "report",
          description: "",
          async execute(content, ctx) {
            await ctx.emit(content, { content });
          },
        });
    },
  }));
  const tid = await berry.create({ extensions: { reports: {}, gate: {} }, model });
  const conversation = (await berry.conversation(tid))!;
  const [before] = await turn(harness, tid, []);
  const previous = (await conversation.context(context)).messages;
  const calls = faux.state.callCount;
  await berry.command(tid, "report", "first edit");
  assertEquals(faux.state.callCount, calls);
  const stored = (await conversation.context(context)).messages;
  assertEquals(stored.slice(0, previous.length), previous);
  assertEquals(stored.at(-1)?.content, "first edit");

  const [after] = await turn(harness, tid, []);
  assertEquals(after.slice(0, stored.length), stored);
  assertEquals(after.filter((m) => m.role === "system"), before.filter((m) => m.role === "system"));

  let next: readonly Message[] = [];
  faux.setResponses([
    fauxAssistantMessage(fauxToolCall("hold", {}), { stopReason: "toolUse" }),
    (request) => (next = request.messages, fauxAssistantMessage("done")),
  ]);
  const running = await conversation.submit({ type: "input", content: "go" }, context);
  await gate.entered.promise;
  const reporting = berry.command(tid, "report", "second edit");
  try {
    await until(async () =>
      (await berry.harness.inspect(context)).submissions.some((s) =>
        s.conversationId === conversation.id && s.type === "write"
      )
    );
    assertEquals(await berry.state(tid, "reports"), ["first edit"]);
    assert(
      !(await conversation.context(context)).messages.some((m) => m.content === "second edit"),
    );
  } finally {
    gate.release.resolve("released");
  }
  await reporting;
  await running.wait(context);
  await conversation.waitForIdle(context);
  assertEquals(await berry.state(tid, "reports"), ["first edit", "second edit"]);
  assertEquals(next.slice(0, stored.length), stored);
  assertEquals(next.at(-1)?.content, "second edit");
  assert(next.findLastIndex((m) => m.role === "toolResult") < next.length - 1);
});

Deno.test("a failed script's images reach anthropic in a result it accepts, and the history keeps the failure", async () => {
  await using harness = await setup(undefined, {
    api: "anthropic-messages",
    models: [{ id: "claude", input: ["text", "image"] }],
  });
  const { berry, model } = harness;
  const png =
    "iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAAC1HAwCAAAAC0lEQVR42mNk+A8AAQUBAScY42YAAAAASUVORK5CYII=";
  const tid = await berry.create({ extensions: {}, model });
  const [, sent] = await turn(harness, tid, [
    script(`
        image({ type: "image", data: "${png}", mimeType: "image/png" });
        throw new Error("after the screenshot");
      `),
  ]);
  const result = sent.findLast((message) => message.role === "toolResult");
  assert(result?.role === "toolResult");
  assertEquals(result.isError, false);
  assertEquals(images(sent), [png]);
  assertStringIncludes(lastToolResult(sent).text, "after the screenshot");
  const stored = await (await berry.conversation(tid))!.context(context);
  const kept = stored.messages.findLast((message) => message.role === "toolResult");
  assert(kept?.role === "toolResult");
  assertEquals(kept.isError, true);
});