Files
harness-tools/extensions/todo-herdr/tests/tools.test.ts
T

339 lines
13 KiB
TypeScript

import assert from "node:assert/strict";
import test from "node:test";
import { TaskState } from "../src/state.ts";
const { validateToolArguments } = await import(
new URL("./utils/validation.js", import.meta.resolve("@earendil-works/pi-ai")).href
);
import { registerTaskTools } from "../src/tools.ts";
function registeredTools(state = new TaskState()) {
const tools = new Map<string, any>();
let changes = 0;
const autoClearRestarts: boolean[] = [];
const pi = {
registerTool(tool: any) {
tools.set(tool.name, tool);
},
};
registerTaskTools(pi as never, state, (_ctx, restartAutoClear = false) => {
changes += 1;
autoClearRestarts.push(restartAutoClear);
});
return { tools, state, changes: () => changes, autoClearRestarts };
}
async function execute(tool: any, params: unknown) {
const prepared = tool.prepareArguments?.(params) ?? params;
const validated = validateToolArguments(tool, { name: tool.name, arguments: prepared } as never);
return tool.execute("call", validated, undefined, undefined, {});
}
test("registers five focused tools without the old add_task alias", () => {
const { tools } = registeredTools();
assert.deepEqual([...tools.keys()], ["set_tasks", "patch_tasks", "rm_task", "list_task", "get_task"]);
assert.deepEqual(Object.keys(tools.get("set_tasks").parameters.properties), ["mode", "tasks", "force"]);
assert.deepEqual(Object.keys(tools.get("patch_tasks").parameters.properties), ["tasks"]);
assert.deepEqual(Object.keys(tools.get("rm_task").parameters.properties), ["ids", "reason"]);
assert.deepEqual(Object.keys(tools.get("list_task").parameters.properties), []);
assert.deepEqual(Object.keys(tools.get("get_task").parameters.properties), ["ids"]);
assert.match(tools.get("patch_tasks").description, /do not recreate cleared completed tasks/);
});
test("set_tasks schema hoists $defs to the root with #/$defs/ refs", () => {
const { tools } = registeredTools();
const params = tools.get("set_tasks").parameters;
const itemSchema = params.properties.tasks.items;
// $defs must live at the document root: root-style refs resolve from there,
// and moonshot-flavored validators reject nested definitions (Console Go).
assert.ok(params.$defs.AddTask, "$defs hoisted to the parameter schema root");
assert.deepEqual(params.required, ["mode", "tasks", "force"]);
assert.deepEqual(params.$defs.AddTask.required, ["name", "description", "status", "parentId", "children"]);
assert.equal(itemSchema.$defs, undefined, "no $defs nested inside tasks items");
assert.equal(itemSchema.$ref, "#/$defs/AddTask");
assert.equal(params.$defs.AddTask.properties.children.items.$ref, "#/$defs/AddTask");
assert.equal(params.$defs.AddTask.$id, undefined, "synthetic $id stripped");
const serialized = JSON.stringify(params);
assert.ok(!/"\$ref":"(?!\#)/.test(serialized), "no bare $ref values remain");
});
test("task schemas publish text and batch limits", () => {
const { tools } = registeredTools();
const setParams = tools.get("set_tasks").parameters;
const addTask = setParams.$defs.AddTask;
const updateTask = tools.get("patch_tasks").parameters.properties.tasks.items;
assert.equal(setParams.properties.tasks.maxItems, 50);
assert.equal(addTask.properties.children.maxItems, 50);
assert.equal(addTask.properties.name.maxLength, 200);
assert.match(addTask.properties.name.pattern, /061C/i);
assert.equal(addTask.properties.description.anyOf[0].maxLength, 4_000);
assert.match(addTask.properties.description.anyOf[0].pattern, /061C/i);
assert.equal(tools.get("patch_tasks").parameters.properties.tasks.maxItems, 50);
assert.equal(updateTask.properties.name.anyOf[0].maxLength, 200);
assert.equal(updateTask.properties.description.anyOf[0].maxLength, 4_000);
assert.equal(tools.get("rm_task").parameters.properties.ids.maxItems, 50);
assert.equal(tools.get("get_task").parameters.properties.ids.maxItems, 50);
const setTool = tools.get("set_tasks");
const validUnicode = setTool.prepareArguments({
mode: "append",
tasks: [{ name: "😀".repeat(200), description: "line one\nline two" }],
});
assert.doesNotThrow(() =>
validateToolArguments(setTool, { name: setTool.name, arguments: validUnicode } as never),
);
const unsafeText = setTool.prepareArguments({ mode: "append", tasks: [{ name: "unsafe\u061cname" }] });
assert.throws(
() => validateToolArguments(setTool, { name: setTool.name, arguments: unsafeText } as never),
/validation failed/i,
);
});
test("set_tasks preparation bounds hostile recursive input before validation", () => {
const { tools } = registeredTools();
const setTool = tools.get("set_tasks");
let nested: Record<string, unknown> = { name: "Leaf" };
for (let depth = 0; depth < 10_000; depth += 1) nested = { name: `Level ${depth}`, children: [nested] };
const prepared = setTool.prepareArguments({ mode: "append", tasks: [nested] });
assert.throws(
() => validateToolArguments(setTool, { name: setTool.name, arguments: prepared } as never),
/validation failed/i,
);
const oversized = setTool.prepareArguments({
mode: "append",
tasks: Array.from({ length: 1_000 }, (_, index) => ({ name: `Task ${index}` })),
});
assert.equal(oversized.tasks.length, 51);
assert.throws(
() => validateToolArguments(setTool, { name: setTool.name, arguments: oversized } as never),
/validation failed/i,
);
const updateTool = tools.get("patch_tasks");
const oversizedUpdate = updateTool.prepareArguments({
tasks: Array.from({ length: 1_000 }, (_, index) => ({ id: index + 1 })),
});
assert.equal(oversizedUpdate.tasks.length, 51);
assert.throws(
() => validateToolArguments(updateTool, { name: updateTool.name, arguments: oversizedUpdate } as never),
/validation failed/i,
);
});
test("strict task schemas normalize legacy optional arguments", () => {
const { tools } = registeredTools();
const setTool = tools.get("set_tasks");
const updateTool = tools.get("patch_tasks");
assert.deepEqual(updateTool.parameters.properties.tasks.items.required, [
"id",
"name",
"description",
"status",
"parentId",
"position",
]);
assert.deepEqual(setTool.prepareArguments({
mode: "append",
tasks: [{ name: "Parent", children: [{ name: "Child" }] }],
}), {
mode: "append",
tasks: [
{
name: "Parent",
description: null,
status: "pending",
parentId: null,
children: [
{
name: "Child",
description: null,
status: "pending",
parentId: null,
children: [],
},
],
},
],
force: false,
});
assert.deepEqual(updateTool.prepareArguments({ tasks: [{ id: 1, status: "done" }] }), {
tasks: [
{
id: 1,
name: null,
description: null,
status: "done",
parentId: -1,
position: -1,
},
],
});
assert.equal(updateTool.prepareArguments({ tasks: [{ id: 1, parentId: null }] }).tasks[0].parentId, 0);
const validated = validateToolArguments(updateTool, {
name: "patch_tasks",
arguments: updateTool.prepareArguments({ tasks: [{ id: 1, status: "done" }] }),
} as never);
assert.equal(validated.tasks[0].parentId, -1);
assert.equal(validated.tasks[0].position, -1);
});
test("validated update sentinels preserve order and map legacy null parent to root", async () => {
const harness = registeredTools();
await execute(harness.tools.get("set_tasks"), {
mode: "append",
tasks: [
{ name: "Parent", children: [{ name: "Child" }] },
{ name: "Sibling" },
],
});
await execute(harness.tools.get("patch_tasks"), { tasks: [{ id: 3, status: "blocked" }] });
assert.deepEqual(
harness.state.list().filter((task) => task.parentId === null).map((task) => task.id),
[1, 3],
);
await execute(harness.tools.get("patch_tasks"), { tasks: [{ id: 2, parentId: null }] });
assert.equal(harness.state.get([2])[0]?.parentId, null);
});
test("patch_tasks explains an auto-cleared empty tree without recreating tasks", async () => {
const harness = registeredTools();
await assert.rejects(
execute(harness.tools.get("patch_tasks"), { tasks: [{ id: 1, status: "done" }] }),
/Patch failed · 1 validation error\n- task tree is empty and may have auto-cleared after completion; do not recreate completed tasks\. Use set_tasks only for genuinely new or remaining work/,
);
assert.deepEqual(harness.state.list(), []);
assert.equal(harness.changes(), 0);
});
test("set_tasks appends optional-description tasks and persists a complete snapshot", async () => {
const harness = registeredTools();
const result = await execute(harness.tools.get("set_tasks"), {
mode: "append",
tasks: [{ name: "Parent", children: [{ name: "Child", description: "Useful detail" }] }],
});
assert.equal(result.content[0].text, "Added 2 tasks: 1, 2. 0/2 done.");
assert.equal(result.details.action, "append");
assert.equal(result.details.snapshot.tasks.length, 2);
assert.equal(result.details.snapshot.tasks[0].description, undefined);
assert.equal(result.details.snapshot.nextId, 3);
assert.equal(harness.changes(), 1);
assert.deepEqual(harness.autoClearRestarts, [false]);
});
test("set_tasks atomically replaces or clears the tree and resets IDs", async () => {
const harness = registeredTools();
await execute(harness.tools.get("set_tasks"), {
mode: "append",
tasks: [{ name: "Old", status: "done" }],
});
const replaced = await execute(harness.tools.get("set_tasks"), {
mode: "replace",
tasks: [{ name: "New", status: "in-progress" }],
});
assert.equal(replaced.content[0].text, "Replaced task tree with 1 task. 0/1 done. Task IDs restart at 1.");
assert.deepEqual(harness.state.list().map((task) => [task.id, task.name]), [[1, "New"]]);
const cleared = await execute(harness.tools.get("set_tasks"), {
mode: "replace",
tasks: [],
force: true,
});
assert.equal(cleared.content[0].text, "Cleared task tree. Task IDs reset to 1.");
assert.deepEqual(harness.state.snapshot(), { tasks: [], nextId: 1 });
assert.deepEqual(harness.autoClearRestarts, [false, true, true]);
});
test("tool validation errors preserve atomic state", async () => {
const harness = registeredTools();
await execute(harness.tools.get("set_tasks"), {
mode: "append",
tasks: [{ name: "Parent", children: [{ name: "Child" }] }],
});
const before = harness.state.snapshot();
await assert.rejects(
execute(harness.tools.get("patch_tasks"), { tasks: [{ id: 1, parentId: 2 }] }),
/Patch failed · 1 validation error\n- task [12]: parentId creates a cycle/,
);
assert.deepEqual(harness.state.snapshot(), before);
assert.equal(harness.changes(), 1);
});
test("rm_task requires and reports a constrained removal reason", async () => {
const harness = registeredTools();
await execute(harness.tools.get("set_tasks"), {
mode: "append",
tasks: [{ name: "Duplicate" }],
});
const removed = await execute(harness.tools.get("rm_task"), { ids: [1], reason: "duplicate" });
assert.match(removed.content[0].text, /Reason: duplicate\.$/);
assert.deepEqual(harness.state.list(), []);
});
test("list is compact and get returns available details for multiple IDs", async () => {
const harness = registeredTools();
await execute(harness.tools.get("set_tasks"), {
mode: "append",
tasks: [{ name: "Parent", description: "Detailed", children: [{ name: "Child" }] }],
});
const list = await execute(harness.tools.get("list_task"), {});
assert.equal(list.content[0].text, "1 [pending] Parent\n└─ 2 [pending] Child");
const get = await execute(harness.tools.get("get_task"), { ids: [1, 2] });
const parsed = JSON.parse(get.content[0].text);
assert.equal(parsed[0].description, "Detailed");
assert.deepEqual(parsed[0].childrenIds, [2]);
assert.equal(parsed[1].description, undefined);
assert.equal(parsed[1].parentId, 1);
});
test("get_task exposes cumulative timing while list output stays compact", async () => {
const startedAt = Date.UTC(2026, 0, 1);
let now = startedAt;
const harness = registeredTools(new TaskState(() => now));
const added = await execute(harness.tools.get("set_tasks"), {
mode: "append",
tasks: [{ name: "Timed", status: "in-progress" }],
});
assert.equal(added.content[0].text, "Added 1 task: 1. 0/1 done.");
assert.equal(added.details.snapshot.timings[0].activeStartedAt, startedAt);
now += 65_432;
const list = await execute(harness.tools.get("list_task"), {});
assert.equal(list.content[0].text, "1 [in-progress] Timed");
const active = JSON.parse((await execute(harness.tools.get("get_task"), { ids: [1] })).content[0].text);
assert.deepEqual(active[0], {
id: 1,
name: "Timed",
status: "in-progress",
parentId: null,
childrenIds: [],
startedAt: new Date(startedAt).toISOString(),
durationMs: 65_432,
});
now += 568;
const updated = await execute(harness.tools.get("patch_tasks"), {
tasks: [{ id: 1, status: "blocked" }],
});
assert.equal(updated.content[0].text, "Updated 1 task: 1. 0/1 done.");
now += 10_000;
const paused = JSON.parse((await execute(harness.tools.get("get_task"), { ids: [1] })).content[0].text);
assert.equal(paused[0].startedAt, new Date(startedAt).toISOString());
assert.equal(paused[0].stoppedAt, new Date(startedAt + 66_000).toISOString());
assert.equal(paused[0].durationMs, 66_000);
});