274 lines
8.9 KiB
TypeScript
274 lines
8.9 KiB
TypeScript
import { afterEach, describe, expect, it, vi } from "vitest";
|
|
import { buildSystemPrompt, buildTaskPrompt } from "../../src/main/prompt-builder/index";
|
|
import type { PromptContext } from "../../src/main/prompt-builder/index";
|
|
|
|
function buildCtx(overrides: Partial<PromptContext> = {}): PromptContext {
|
|
return {
|
|
agent: {
|
|
id: "agent-1",
|
|
companyId: "company-1",
|
|
name: "CTO",
|
|
role: "CTO",
|
|
title: "Chief Technology Officer",
|
|
roleLevel: "c_suite",
|
|
department: "engineering",
|
|
status: "active",
|
|
reportsTo: "ceo-1",
|
|
connectorId: "openai-codex",
|
|
workspaceId: null,
|
|
model: null,
|
|
capabilities: "Lead engineering",
|
|
budgetMonthlyUsd: 200,
|
|
spentMonthlyUsd: 0,
|
|
metadataJson: "{}",
|
|
heartbeatEnabled: true,
|
|
heartbeatIntervalSec: 120,
|
|
lastHeartbeatAt: null,
|
|
createdAt: new Date().toISOString(),
|
|
updatedAt: new Date().toISOString(),
|
|
},
|
|
apiUrl: "http://127.0.0.1:3210",
|
|
apiKey: "token",
|
|
runId: "run-1",
|
|
task: null,
|
|
companyName: "Prompt Co",
|
|
companyDescription: "Tests leadership prompt classification",
|
|
goals: [],
|
|
projects: [],
|
|
directReports: [],
|
|
chainOfCommand: [{
|
|
id: "ceo-1",
|
|
companyId: "company-1",
|
|
name: "CEO",
|
|
role: "CEO",
|
|
title: "Chief Executive Officer",
|
|
roleLevel: "c_suite",
|
|
department: "executive",
|
|
status: "active",
|
|
reportsTo: null,
|
|
connectorId: "openai-codex",
|
|
workspaceId: null,
|
|
model: null,
|
|
capabilities: "Lead the company",
|
|
budgetMonthlyUsd: 250,
|
|
spentMonthlyUsd: 0,
|
|
metadataJson: "{}",
|
|
heartbeatEnabled: true,
|
|
heartbeatIntervalSec: 120,
|
|
lastHeartbeatAt: null,
|
|
createdAt: new Date().toISOString(),
|
|
updatedAt: new Date().toISOString(),
|
|
}],
|
|
assignedTasks: [],
|
|
allAgents: [],
|
|
pendingApprovals: [],
|
|
tasksAwaitingReview: [],
|
|
wakeReason: null,
|
|
allTasks: [],
|
|
recentComments: [],
|
|
socialAccounts: [],
|
|
...overrides,
|
|
};
|
|
}
|
|
|
|
function buildTask(overrides: Partial<NonNullable<PromptContext["task"]>> = {}): NonNullable<PromptContext["task"]> {
|
|
const now = new Date().toISOString();
|
|
return {
|
|
id: "task-1",
|
|
companyId: "company-1",
|
|
projectId: null,
|
|
goalId: null,
|
|
parentId: null,
|
|
title: "Ship volatile task",
|
|
description: "This task should live in the volatile prompt tail.",
|
|
assigneeAgentId: "agent-1",
|
|
workspaceId: null,
|
|
status: "in_progress",
|
|
priority: "high",
|
|
activeRunId: null,
|
|
costUsd: 0,
|
|
taskType: null,
|
|
requiresUserReview: false,
|
|
requiresFilesystem: false,
|
|
startedAt: null,
|
|
completedAt: null,
|
|
cancelledAt: null,
|
|
createdAt: now,
|
|
updatedAt: now,
|
|
...overrides,
|
|
};
|
|
}
|
|
|
|
function buildApproval(overrides: Partial<PromptContext["pendingApprovals"][number]> = {}): PromptContext["pendingApprovals"][number] {
|
|
const now = new Date().toISOString();
|
|
return {
|
|
id: "approval-1",
|
|
companyId: "company-1",
|
|
relatedTaskId: null,
|
|
relatedAgentId: null,
|
|
requestedByAgentId: "agent-1",
|
|
type: "approve_ceo_strategy",
|
|
payloadSummary: "Review the volatile org gate.",
|
|
impactSummary: "Board decision required.",
|
|
payloadJson: null,
|
|
decisionNote: "",
|
|
state: "pending",
|
|
createdAt: now,
|
|
updatedAt: now,
|
|
...overrides,
|
|
};
|
|
}
|
|
|
|
describe("buildSystemPrompt leadership classification", () => {
|
|
afterEach(() => {
|
|
vi.useRealTimers();
|
|
});
|
|
|
|
it("treats first-layer executive leaders as managers even before they have direct reports", () => {
|
|
const prompt = buildSystemPrompt(buildCtx());
|
|
|
|
expect(prompt).toContain("Operating Instructions (Manager)");
|
|
expect(prompt).toContain("New hires for your team should report to you by default");
|
|
expect(prompt).not.toContain("Operating Instructions (Individual Contributor)");
|
|
});
|
|
|
|
it("renders projected skills without pi skill shape casts", () => {
|
|
const prompt = buildSystemPrompt(buildCtx({
|
|
skills: [{
|
|
name: "ui<review>&qa",
|
|
description: "Use for A&B \"checks\" and 'snapshots'.",
|
|
filePath: "/tmp/skills/ui<review>&qa/SKILL.md",
|
|
baseDir: "/tmp/skills/ui<review>&qa",
|
|
}],
|
|
}));
|
|
|
|
expect(prompt).toContain("<available_skills>");
|
|
expect(prompt).toContain("<name>ui<review>&qa</name>");
|
|
expect(prompt).toContain("<description>Use for A&B "checks" and 'snapshots'.</description>");
|
|
expect(prompt).toContain("<location>/tmp/skills/ui<review>&qa/SKILL.md</location>");
|
|
});
|
|
|
|
it("keeps the system prompt byte-stable across wall-clock changes", () => {
|
|
const ctx = buildCtx();
|
|
|
|
vi.useFakeTimers();
|
|
vi.setSystemTime(new Date("2026-04-22T10:00:00.000Z"));
|
|
const first = buildSystemPrompt(ctx);
|
|
vi.setSystemTime(new Date("2026-04-23T23:59:00.000Z"));
|
|
const second = buildSystemPrompt(ctx);
|
|
|
|
expect(second).toBe(first);
|
|
});
|
|
|
|
it("keeps run, locale, and task context out of the stable system prompt", () => {
|
|
const ctx = buildCtx({
|
|
runId: "run-volatile",
|
|
locale: "zh",
|
|
task: buildTask(),
|
|
wakeReason: "assignment",
|
|
});
|
|
|
|
const stablePrompt = buildSystemPrompt(ctx);
|
|
const fullPrompt = buildTaskPrompt(ctx);
|
|
|
|
expect(stablePrompt).not.toContain("run-volatile");
|
|
expect(stablePrompt).not.toContain("Ship volatile task");
|
|
expect(stablePrompt).not.toContain("Respond to the user in Chinese.");
|
|
expect(stablePrompt).not.toContain("{{runId}}");
|
|
expect(stablePrompt).toContain("$AGENT_COMPANY_RUN_ID");
|
|
expect(fullPrompt).toContain("Current Run ID: run-volatile");
|
|
expect(fullPrompt).toContain("Ship volatile task");
|
|
expect(fullPrompt).toContain("Respond to the user in Chinese.");
|
|
});
|
|
|
|
it("keeps the stable prefix identical across volatile run changes", () => {
|
|
const firstCtx = buildCtx({
|
|
runId: "run-a",
|
|
locale: "en",
|
|
task: buildTask({ id: "task-a", title: "First volatile task" }),
|
|
wakeReason: "timer",
|
|
});
|
|
const secondCtx = buildCtx({
|
|
runId: "run-b",
|
|
locale: "zh",
|
|
task: buildTask({ id: "task-b", title: "Second volatile task" }),
|
|
wakeReason: "assignment",
|
|
});
|
|
|
|
const firstStable = buildSystemPrompt(firstCtx);
|
|
const secondStable = buildSystemPrompt(secondCtx);
|
|
const secondFull = buildTaskPrompt(secondCtx);
|
|
|
|
expect(secondStable).toBe(firstStable);
|
|
expect(secondFull.slice(0, secondStable.length)).toBe(secondStable);
|
|
expect(secondFull.slice(secondStable.length)).toContain("Second volatile task");
|
|
});
|
|
|
|
it("keeps approval gates and workload counts out of the stable system prompt", () => {
|
|
const baseAgent = buildCtx().agent;
|
|
const report = {
|
|
...baseAgent,
|
|
id: "report-1",
|
|
name: "Engineer",
|
|
role: "Engineer",
|
|
title: "Staff Engineer",
|
|
reportsTo: baseAgent.id,
|
|
};
|
|
const stableBase = buildCtx({
|
|
chainOfCommand: [],
|
|
directReports: [report],
|
|
allAgents: [baseAgent, report],
|
|
allTasks: [buildTask({ id: "report-task", assigneeAgentId: report.id, status: "todo" })],
|
|
pendingApprovals: [],
|
|
});
|
|
const volatileChanged = buildCtx({
|
|
chainOfCommand: [],
|
|
directReports: [report],
|
|
allAgents: [baseAgent, report],
|
|
allTasks: [buildTask({ id: "report-task", assigneeAgentId: report.id, status: "blocked" })],
|
|
pendingApprovals: [buildApproval()],
|
|
});
|
|
|
|
const stablePrompt = buildSystemPrompt(stableBase);
|
|
const changedStablePrompt = buildSystemPrompt(volatileChanged);
|
|
const changedFullPrompt = buildTaskPrompt(volatileChanged);
|
|
|
|
expect(changedStablePrompt).toBe(stablePrompt);
|
|
expect(changedFullPrompt).toContain("Pending Approvals");
|
|
expect(changedFullPrompt).toContain("1 blocked");
|
|
});
|
|
|
|
it("keeps heartbeat skip state out of the task prompt", () => {
|
|
const ctx = buildCtx() as PromptContext & {
|
|
heartbeatSkip: { reason: string; streak: number; since: string; lastTickAt: string };
|
|
};
|
|
ctx.heartbeatSkip = {
|
|
reason: "worker_not_ready",
|
|
streak: 7,
|
|
since: "2026-04-23T00:00:00.000Z",
|
|
lastTickAt: "2026-04-23T00:10:00.000Z",
|
|
};
|
|
|
|
const prompt = buildTaskPrompt(ctx);
|
|
|
|
expect(prompt).not.toContain("Heartbeat Skip Status");
|
|
expect(prompt).not.toContain("worker_not_ready");
|
|
expect(prompt).not.toContain("Consecutive skips");
|
|
});
|
|
|
|
it("keeps volatile task context after the cacheable system prefix", () => {
|
|
const ctx = buildCtx({ wakeReason: "timer" });
|
|
const systemPrompt = buildSystemPrompt(ctx);
|
|
|
|
vi.useFakeTimers();
|
|
vi.setSystemTime(new Date("2026-04-22T10:00:00.000Z"));
|
|
const first = buildTaskPrompt(ctx);
|
|
vi.setSystemTime(new Date("2026-04-23T23:59:00.000Z"));
|
|
const second = buildTaskPrompt(ctx);
|
|
|
|
expect(first.slice(0, systemPrompt.length)).toBe(systemPrompt);
|
|
expect(second.slice(0, systemPrompt.length)).toBe(systemPrompt);
|
|
expect(first.slice(systemPrompt.length)).not.toBe(second.slice(systemPrompt.length));
|
|
});
|
|
});
|