can1357
1288 lines
43 KiB
TypeScript
1288 lines
43 KiB
TypeScript
import { afterAll, afterEach, beforeAll, beforeEach, describe, expect, it, vi } from "bun:test";
|
|
import * as path from "node:path";
|
|
import { Agent, type AgentMessage } from "@oh-my-pi/pi-agent-core";
|
|
import * as compactionModule from "@oh-my-pi/pi-agent-core/compaction";
|
|
import type { AssistantMessage, Model, ToolCall } from "@oh-my-pi/pi-ai";
|
|
import { createMockModel } from "@oh-my-pi/pi-ai/providers/mock";
|
|
import { getBundledModel } from "@oh-my-pi/pi-catalog/models";
|
|
import { ModelRegistry } from "@oh-my-pi/pi-coding-agent/config/model-registry";
|
|
import { Settings } from "@oh-my-pi/pi-coding-agent/config/settings";
|
|
import { ExtensionRunner, loadExtensions } from "@oh-my-pi/pi-coding-agent/extensibility/extensions";
|
|
import { SecretObfuscator } from "@oh-my-pi/pi-coding-agent/secrets";
|
|
import { AgentSession, type AgentSessionEvent } from "@oh-my-pi/pi-coding-agent/session/agent-session";
|
|
import { AuthStorage } from "@oh-my-pi/pi-coding-agent/session/auth-storage";
|
|
import { SessionManager } from "@oh-my-pi/pi-coding-agent/session/session-manager";
|
|
import { TempDir } from "@oh-my-pi/pi-utils";
|
|
|
|
const HANDOFF_SECRET = "HANDOFF_SECRET_TOKEN_12345";
|
|
|
|
describe("AgentSession handoff", () => {
|
|
// Immutable across the whole file: the model registry's synchronous bundled-model
|
|
// load dominates per-test setup (~100ms each), and the auth store + bundled model
|
|
// never change. Build them once. Per-test mutable state (session, session file,
|
|
// emitted events) is rebuilt in beforeEach.
|
|
let sharedDir: TempDir;
|
|
let authStorage: AuthStorage;
|
|
let modelRegistry: ModelRegistry;
|
|
let model: Model;
|
|
|
|
let tempDir: TempDir;
|
|
let session: AgentSession;
|
|
let sessionManager: SessionManager;
|
|
let events: AgentSessionEvent[];
|
|
let obfuscator: SecretObfuscator;
|
|
|
|
/** Poll `predicate` until it holds (returns as soon as the state is reached) or the
|
|
* deadline elapses. Replaces blind settle sleeps for tests with a positive signal. */
|
|
async function waitFor(predicate: () => boolean, timeoutMs = 1_000): Promise<void> {
|
|
const deadline = Date.now() + timeoutMs;
|
|
while (!predicate()) {
|
|
if (Date.now() >= deadline) {
|
|
throw new Error("Timed out waiting for condition");
|
|
}
|
|
await Bun.sleep(1);
|
|
}
|
|
}
|
|
|
|
/** Drain post-turn maintenance deterministically for negative tests (those proving
|
|
* maintenance did NOT run, where there is no positive signal to poll on). Post-turn
|
|
* work is scheduled fire-and-forget: a single event-loop turn lets the handler run to
|
|
* its decision and register any compaction pass as a tracked post-prompt task, then
|
|
* `waitForIdle()` drains that task to completion. */
|
|
async function drainMaintenance(): Promise<void> {
|
|
await Bun.sleep(0);
|
|
await session.waitForIdle();
|
|
}
|
|
|
|
beforeAll(async () => {
|
|
sharedDir = TempDir.createSync("@pi-handoff-shared-");
|
|
authStorage = await AuthStorage.create(path.join(sharedDir.path(), "testauth.db"));
|
|
authStorage.setRuntimeApiKey("anthropic", "test-key");
|
|
modelRegistry = new ModelRegistry(authStorage);
|
|
|
|
const bundled = getBundledModel("anthropic", "claude-sonnet-4-5");
|
|
if (!bundled) {
|
|
throw new Error("Expected built-in anthropic model to exist");
|
|
}
|
|
model = bundled;
|
|
});
|
|
|
|
afterAll(async () => {
|
|
authStorage.close();
|
|
try {
|
|
await sharedDir.remove();
|
|
} catch {}
|
|
});
|
|
|
|
beforeEach(async () => {
|
|
tempDir = TempDir.createSync("@pi-handoff-");
|
|
sessionManager = SessionManager.create(tempDir.path(), tempDir.path());
|
|
events = [];
|
|
obfuscator = new SecretObfuscator([{ type: "plain", content: HANDOFF_SECRET }]);
|
|
|
|
const agent = new Agent({
|
|
initialState: {
|
|
model,
|
|
systemPrompt: ["Test"],
|
|
tools: [],
|
|
messages: [],
|
|
},
|
|
});
|
|
|
|
session = new AgentSession({
|
|
agent,
|
|
sessionManager,
|
|
settings: Settings.isolated({
|
|
"compaction.enabled": true,
|
|
"compaction.autoContinue": false,
|
|
}),
|
|
modelRegistry,
|
|
obfuscator,
|
|
});
|
|
|
|
session.subscribe(event => {
|
|
events.push(event);
|
|
});
|
|
|
|
sessionManager.appendMessage({
|
|
role: "user",
|
|
content: [{ type: "text", text: "seed" }],
|
|
timestamp: Date.now() - 2,
|
|
});
|
|
sessionManager.appendMessage({
|
|
role: "assistant",
|
|
content: [{ type: "text", text: "seed response" }],
|
|
api: model.api,
|
|
provider: model.provider,
|
|
model: model.id,
|
|
stopReason: "stop",
|
|
usage: {
|
|
input: 16,
|
|
output: 8,
|
|
cacheRead: 0,
|
|
cacheWrite: 0,
|
|
totalTokens: 24,
|
|
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
|
|
},
|
|
timestamp: Date.now() - 1,
|
|
});
|
|
});
|
|
|
|
afterEach(async () => {
|
|
if (session) {
|
|
await session.dispose();
|
|
}
|
|
try {
|
|
await tempDir.remove();
|
|
} catch {}
|
|
vi.restoreAllMocks();
|
|
});
|
|
|
|
it("does not run auto-compaction after handoff turn completes", async () => {
|
|
const handoffText = "## Goal\nContinue from here";
|
|
const generateHandoffSpy = vi.spyOn(compactionModule, "generateHandoff").mockResolvedValue(handoffText);
|
|
|
|
const result = await session.handoff();
|
|
await drainMaintenance();
|
|
|
|
expect(generateHandoffSpy).toHaveBeenCalledTimes(1);
|
|
expect(result?.document).toBe(handoffText);
|
|
|
|
expect(events.filter(event => event.type === "auto_compaction_start")).toHaveLength(0);
|
|
expect(events.filter(event => event.type === "auto_compaction_end")).toHaveLength(0);
|
|
expect(sessionManager.getEntries().filter(entry => entry.type === "compaction")).toHaveLength(0);
|
|
});
|
|
|
|
it("preserves queued steering and follow-up messages across the handoff reset", async () => {
|
|
// Defect 2: handoff() calls agent.reset(), which clears the core steering/follow-up
|
|
// queues. Steers/follow-ups already queued (the mis-routed first compaction message,
|
|
// or RPC/SDK steer()/followUp() issued during the handoff) must survive into the new
|
|
// session instead of being silently dropped.
|
|
vi.spyOn(compactionModule, "generateHandoff").mockResolvedValue("## Goal\nContinue");
|
|
|
|
const textOf = (message: AgentMessage): string => {
|
|
if (!("content" in message)) return "";
|
|
const content = message.content;
|
|
if (typeof content === "string") return content;
|
|
const textBlock = content.find(block => block.type === "text");
|
|
return textBlock?.type === "text" ? textBlock.text : "";
|
|
};
|
|
|
|
const userMsg: AgentMessage = {
|
|
role: "user",
|
|
content: [{ type: "text", text: "keep-steer" }],
|
|
attribution: "user",
|
|
timestamp: Date.now(),
|
|
};
|
|
// A hidden, user-attributed companion (e.g. an ultrathink notice). It is
|
|
// display:false, so isUserQueuedMessage(...) is false for it: preservation must
|
|
// keep it adjacent to its prompt rather than filter it out or reorder it.
|
|
const companionMsg: AgentMessage = {
|
|
role: "custom",
|
|
customType: "ultrathink-notice",
|
|
content: [{ type: "text", text: "companion" }],
|
|
attribution: "user",
|
|
display: false,
|
|
timestamp: Date.now(),
|
|
};
|
|
const followUpMsg: AgentMessage = {
|
|
role: "user",
|
|
content: [{ type: "text", text: "keep-followup" }],
|
|
attribution: "user",
|
|
timestamp: Date.now(),
|
|
};
|
|
session.agent.steer(userMsg);
|
|
session.agent.steer(companionMsg);
|
|
session.agent.followUp(followUpMsg);
|
|
expect(session.agent.hasQueuedMessages()).toBe(true);
|
|
|
|
await session.handoff();
|
|
|
|
expect(session.agent.peekSteeringQueue().map(textOf)).toEqual(["keep-steer", "companion"]);
|
|
expect(session.agent.peekFollowUpQueue().map(textOf)).toEqual(["keep-followup"]);
|
|
});
|
|
|
|
it("preserves steering and follow-up messages enqueued while the handoff is in flight", async () => {
|
|
// Defect 2 in-flight window: the queue snapshot must be captured immediately before
|
|
// agent.reset() (after generateHandoff resolves), NOT at handoff entry. A steer or
|
|
// follow-up issued WHILE the handoff document is still generating must survive the
|
|
// reset — proving capture happens late rather than at the start of handoff().
|
|
const { promise: handoffDoc, resolve: releaseHandoff } = Promise.withResolvers<string>();
|
|
let generateHandoffCalled = false;
|
|
vi.spyOn(compactionModule, "generateHandoff").mockImplementation(async () => {
|
|
generateHandoffCalled = true;
|
|
return handoffDoc;
|
|
});
|
|
|
|
const textOf = (message: AgentMessage): string => {
|
|
if (!("content" in message)) return "";
|
|
const content = message.content;
|
|
if (typeof content === "string") return content;
|
|
const textBlock = content.find(block => block.type === "text");
|
|
return textBlock?.type === "text" ? textBlock.text : "";
|
|
};
|
|
|
|
const handoffPromise = session.handoff();
|
|
// Block until we are genuinely mid-handoff (document generation in flight).
|
|
await waitFor(() => generateHandoffCalled);
|
|
|
|
// Enqueue AFTER generation started but BEFORE it resolves — the window where the old
|
|
// session is still live and agent.reset() has not yet fired.
|
|
session.agent.steer({
|
|
role: "user",
|
|
content: [{ type: "text", text: "inflight-steer" }],
|
|
attribution: "user",
|
|
timestamp: Date.now(),
|
|
});
|
|
session.agent.followUp({
|
|
role: "user",
|
|
content: [{ type: "text", text: "inflight-followup" }],
|
|
attribution: "user",
|
|
timestamp: Date.now(),
|
|
});
|
|
|
|
releaseHandoff("## Goal\nContinue");
|
|
await handoffPromise;
|
|
|
|
expect(session.agent.peekSteeringQueue().map(textOf)).toEqual(["inflight-steer"]);
|
|
expect(session.agent.peekFollowUpQueue().map(textOf)).toEqual(["inflight-followup"]);
|
|
});
|
|
|
|
it("obfuscates custom instructions before generating a handoff", async () => {
|
|
const placeholder = obfuscator.obfuscate(HANDOFF_SECRET);
|
|
const generateHandoffSpy = vi
|
|
.spyOn(compactionModule, "generateHandoff")
|
|
.mockResolvedValue(`## Goal\nKeep ${placeholder}`);
|
|
|
|
const result = await session.handoff(`preserve ${HANDOFF_SECRET}`);
|
|
|
|
const handoffCall = generateHandoffSpy.mock.calls[0];
|
|
if (!handoffCall) throw new Error("Expected generateHandoff call");
|
|
expect(handoffCall[3].customInstructions).toBe(`preserve ${placeholder}`);
|
|
expect(handoffCall[3].customInstructions).not.toContain(HANDOFF_SECRET);
|
|
expect(result?.document).toContain(HANDOFF_SECRET);
|
|
expect(result?.document).not.toContain(placeholder);
|
|
});
|
|
|
|
it("obfuscates previous compaction summary and preserve data before forwarding to compact()", async () => {
|
|
const placeholder = obfuscator.obfuscate(HANDOFF_SECRET);
|
|
const entries = sessionManager.getBranch();
|
|
const lastEntryId = entries[entries.length - 1]?.id;
|
|
if (!lastEntryId) throw new Error("Expected a seeded entry id");
|
|
const fixedPreparation: compactionModule.CompactionPreparation = {
|
|
firstKeptEntryId: lastEntryId,
|
|
messagesToSummarize: [{ role: "user", content: [{ type: "text", text: "old" }], timestamp: 1 }],
|
|
turnPrefixMessages: [],
|
|
recentMessages: [],
|
|
isSplitTurn: false,
|
|
tokensBefore: 100,
|
|
previousSummary: `summary ${HANDOFF_SECRET}`,
|
|
previousPreserveData: {
|
|
openaiRemoteCompaction: {
|
|
replacementHistory: [{ role: "user", content: `history ${HANDOFF_SECRET}` }],
|
|
},
|
|
},
|
|
fileOps: { read: new Set(), written: new Set(), edited: new Set() },
|
|
settings: compactionModule.DEFAULT_COMPACTION_SETTINGS,
|
|
};
|
|
vi.spyOn(compactionModule, "prepareCompaction").mockReturnValue(fixedPreparation);
|
|
|
|
const compactSpy = vi.spyOn(compactionModule, "compact").mockResolvedValue({
|
|
summary: "new summary",
|
|
shortSummary: undefined,
|
|
firstKeptEntryId: lastEntryId,
|
|
tokensBefore: 100,
|
|
details: {},
|
|
});
|
|
|
|
await session.compact();
|
|
|
|
const call = compactSpy.mock.calls[0];
|
|
if (!call) throw new Error("Expected compact call");
|
|
expect(call[0].previousSummary).toBe(`summary ${placeholder}`);
|
|
expect(call[0].previousSummary).not.toContain(HANDOFF_SECRET);
|
|
const preserveData = JSON.stringify(call[0].previousPreserveData);
|
|
expect(preserveData).toContain(placeholder);
|
|
expect(preserveData).not.toContain(HANDOFF_SECRET);
|
|
});
|
|
|
|
it("runs context maintenance before sending an oversized pending prompt", async () => {
|
|
session.settings.set("compaction.thresholdTokens", 50);
|
|
session.settings.set("compaction.keepRecentTokens", 1);
|
|
session.settings.set("contextPromotion.enabled", false);
|
|
|
|
const compactSpy = vi.spyOn(compactionModule, "compact").mockImplementation(async preparation => ({
|
|
summary: "pre-prompt compacted",
|
|
shortSummary: undefined,
|
|
firstKeptEntryId: preparation.firstKeptEntryId,
|
|
tokensBefore: preparation.tokensBefore,
|
|
details: {},
|
|
}));
|
|
const promptSpy = vi.spyOn(session.agent, "prompt").mockImplementation(async () => {
|
|
expect(sessionManager.getEntries().some(entry => entry.type === "compaction")).toBe(true);
|
|
});
|
|
|
|
await session.prompt("pending prompt ".repeat(120));
|
|
await waitFor(
|
|
() =>
|
|
compactSpy.mock.calls.length === 1 &&
|
|
events.some(event => event.type === "auto_compaction_end" && event.aborted === false),
|
|
);
|
|
|
|
expect(compactSpy).toHaveBeenCalledTimes(1);
|
|
expect(promptSpy).toHaveBeenCalledTimes(1);
|
|
expect(events).toContainEqual({ type: "auto_compaction_start", reason: "threshold", action: "context-full" });
|
|
expect(events.some(event => event.type === "auto_compaction_end" && event.aborted === false)).toBe(true);
|
|
});
|
|
|
|
it("falls back after one auto-compaction timeout instead of retrying the same model", async () => {
|
|
session.settings.set("compaction.thresholdTokens", 50);
|
|
session.settings.set("compaction.keepRecentTokens", 1);
|
|
session.settings.set("contextPromotion.enabled", false);
|
|
session.settings.set("retry.baseDelayMs", 1);
|
|
|
|
let firstCandidateKey: string | undefined;
|
|
let fallbackCandidateKey: string | undefined;
|
|
const compactSpy = vi.spyOn(compactionModule, "compact").mockImplementation(async (preparation, candidate) => {
|
|
const candidateKey = `${candidate.provider}/${candidate.id}`;
|
|
firstCandidateKey ??= candidateKey;
|
|
if (candidateKey === firstCandidateKey) {
|
|
throw new Error("Summarization failed: The operation timed out.");
|
|
}
|
|
fallbackCandidateKey = candidateKey;
|
|
return {
|
|
summary: "fallback compacted",
|
|
shortSummary: undefined,
|
|
firstKeptEntryId: preparation.firstKeptEntryId,
|
|
tokensBefore: preparation.tokensBefore,
|
|
details: {},
|
|
};
|
|
});
|
|
const promptSpy = vi.spyOn(session.agent, "prompt").mockImplementation(async () => {
|
|
expect(sessionManager.getEntries().some(entry => entry.type === "compaction")).toBe(true);
|
|
});
|
|
|
|
await session.prompt("pending prompt ".repeat(120));
|
|
await waitFor(
|
|
() =>
|
|
fallbackCandidateKey !== undefined &&
|
|
events.some(event => event.type === "auto_compaction_end" && event.aborted === false),
|
|
);
|
|
|
|
expect(
|
|
compactSpy.mock.calls.filter(call => `${call[1].provider}/${call[1].id}` === firstCandidateKey),
|
|
).toHaveLength(1);
|
|
expect(fallbackCandidateKey).toBeDefined();
|
|
expect(promptSpy).toHaveBeenCalledTimes(1);
|
|
});
|
|
it("keeps pre-prompt context-full checks aligned with provider-anchored usage", async () => {
|
|
await session.dispose();
|
|
authStorage.setRuntimeApiKey("openai", "test-key");
|
|
sessionManager = SessionManager.create(tempDir.path(), tempDir.path());
|
|
events = [];
|
|
|
|
const mock = createMockModel({
|
|
id: "gpt-5.5",
|
|
provider: "openai",
|
|
contextWindow: 10_000,
|
|
responses: [
|
|
{
|
|
content: ["ok"],
|
|
stopReason: "stop",
|
|
usage: {
|
|
input: 1_005,
|
|
output: 20,
|
|
cacheRead: 0,
|
|
cacheWrite: 0,
|
|
totalTokens: 1_025,
|
|
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
|
|
},
|
|
},
|
|
],
|
|
});
|
|
const seedUser: AgentMessage = {
|
|
role: "user",
|
|
content: [{ type: "text", text: "seed" }],
|
|
timestamp: Date.now() - 2,
|
|
};
|
|
const seedAssistant: AssistantMessage = {
|
|
role: "assistant",
|
|
content: [
|
|
{
|
|
type: "thinking",
|
|
thinking: "short reasoning",
|
|
thinkingSignature: JSON.stringify({
|
|
id: "rs_repro",
|
|
type: "reasoning",
|
|
content: [],
|
|
encrypted_content: "blob ".repeat(30_000),
|
|
summary: [],
|
|
}),
|
|
},
|
|
{ type: "text", text: "done" },
|
|
],
|
|
api: mock.api,
|
|
provider: "openai",
|
|
model: mock.id,
|
|
stopReason: "stop",
|
|
usage: {
|
|
input: 1_000,
|
|
output: 10,
|
|
cacheRead: 0,
|
|
cacheWrite: 0,
|
|
totalTokens: 1_010,
|
|
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
|
|
},
|
|
timestamp: Date.now() - 1,
|
|
};
|
|
sessionManager.appendMessage(seedUser);
|
|
sessionManager.appendMessage(seedAssistant);
|
|
|
|
const agent = new Agent({
|
|
getApiKey: () => "test-key",
|
|
initialState: {
|
|
model: mock,
|
|
systemPrompt: ["Test"],
|
|
tools: [],
|
|
messages: [seedUser, seedAssistant],
|
|
},
|
|
streamFn: mock.stream,
|
|
});
|
|
session = new AgentSession({
|
|
agent,
|
|
sessionManager,
|
|
settings: Settings.isolated({
|
|
"compaction.enabled": true,
|
|
"compaction.autoContinue": false,
|
|
"compaction.strategy": "context-full",
|
|
"compaction.thresholdTokens": 8_000,
|
|
"contextPromotion.enabled": false,
|
|
}),
|
|
modelRegistry,
|
|
});
|
|
session.subscribe(event => {
|
|
events.push(event);
|
|
});
|
|
const compactSpy = vi.spyOn(compactionModule, "compact").mockImplementation(async preparation => ({
|
|
summary: "pre-prompt compacted",
|
|
shortSummary: undefined,
|
|
firstKeptEntryId: preparation.firstKeptEntryId,
|
|
tokensBefore: preparation.tokensBefore,
|
|
details: {},
|
|
}));
|
|
|
|
expect(session.getContextUsage({ contextWindow: 10_000 })).toMatchObject({
|
|
tokens: 1_000,
|
|
contextWindow: 10_000,
|
|
percent: 10,
|
|
});
|
|
|
|
await session.prompt("small pending prompt");
|
|
|
|
expect(compactSpy).not.toHaveBeenCalled();
|
|
expect(events.filter(event => event.type === "auto_compaction_start")).toHaveLength(0);
|
|
expect(mock.calls).toHaveLength(1);
|
|
});
|
|
it("counts current non-message token growth in provider-anchored pre-prompt checks", async () => {
|
|
await session.dispose();
|
|
authStorage.setRuntimeApiKey("openai", "test-key");
|
|
sessionManager = SessionManager.create(tempDir.path(), tempDir.path());
|
|
events = [];
|
|
|
|
const extensionsResult = await loadExtensions([], tempDir.path());
|
|
const extensionRunner = new ExtensionRunner(
|
|
extensionsResult.extensions,
|
|
extensionsResult.runtime,
|
|
tempDir.path(),
|
|
sessionManager,
|
|
modelRegistry,
|
|
);
|
|
const emitBeforeAgentStart = vi
|
|
.spyOn(extensionRunner, "emitBeforeAgentStart")
|
|
.mockResolvedValueOnce(undefined)
|
|
.mockResolvedValueOnce({ systemPrompt: ["expanded system prompt ".repeat(30_000)] });
|
|
vi.spyOn(extensionRunner, "emit").mockResolvedValue(undefined);
|
|
|
|
const mock = createMockModel({
|
|
id: "gpt-5.5",
|
|
provider: "openai",
|
|
contextWindow: 10_000,
|
|
responses: [
|
|
{
|
|
content: ["seed response"],
|
|
stopReason: "stop",
|
|
usage: {
|
|
input: 1_000,
|
|
output: 10,
|
|
cacheRead: 0,
|
|
cacheWrite: 0,
|
|
totalTokens: 1_010,
|
|
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
|
|
},
|
|
},
|
|
],
|
|
});
|
|
const agent = new Agent({
|
|
getApiKey: () => "test-key",
|
|
initialState: {
|
|
model: mock,
|
|
systemPrompt: ["Test"],
|
|
tools: [],
|
|
messages: [],
|
|
},
|
|
streamFn: mock.stream,
|
|
});
|
|
session = new AgentSession({
|
|
agent,
|
|
sessionManager,
|
|
settings: Settings.isolated({
|
|
"compaction.enabled": true,
|
|
"compaction.autoContinue": false,
|
|
"compaction.strategy": "context-full",
|
|
"compaction.thresholdTokens": 8_000,
|
|
"compaction.keepRecentTokens": 1,
|
|
"contextPromotion.enabled": false,
|
|
}),
|
|
modelRegistry,
|
|
extensionRunner,
|
|
});
|
|
session.subscribe(event => {
|
|
events.push(event);
|
|
});
|
|
|
|
await session.prompt("seed prompt");
|
|
expect(mock.calls).toHaveLength(1);
|
|
expect(session.getContextUsage({ contextWindow: 10_000 })).toMatchObject({
|
|
tokens: 1_000,
|
|
contextWindow: 10_000,
|
|
percent: 10,
|
|
});
|
|
|
|
const compactSpy = vi.spyOn(compactionModule, "compact").mockImplementation(async preparation => ({
|
|
summary: "pre-prompt compacted",
|
|
shortSummary: undefined,
|
|
firstKeptEntryId: preparation.firstKeptEntryId,
|
|
tokensBefore: preparation.tokensBefore,
|
|
details: {},
|
|
}));
|
|
const promptSpy = vi.spyOn(session.agent, "prompt").mockImplementation(async () => {
|
|
expect(sessionManager.getEntries().some(entry => entry.type === "compaction")).toBe(true);
|
|
});
|
|
|
|
await session.prompt("small pending prompt");
|
|
await waitFor(
|
|
() =>
|
|
compactSpy.mock.calls.length === 1 &&
|
|
events.some(event => event.type === "auto_compaction_end" && event.aborted === false),
|
|
);
|
|
|
|
expect(emitBeforeAgentStart).toHaveBeenCalledTimes(2);
|
|
expect(compactSpy).toHaveBeenCalledTimes(1);
|
|
expect(promptSpy).toHaveBeenCalledTimes(1);
|
|
expect(events).toContainEqual({ type: "auto_compaction_start", reason: "threshold", action: "context-full" });
|
|
});
|
|
|
|
it("does not double-count unchanged non-message tokens in provider-anchored pre-prompt checks", async () => {
|
|
await session.dispose();
|
|
authStorage.setRuntimeApiKey("openai", "test-key");
|
|
sessionManager = SessionManager.create(tempDir.path(), tempDir.path());
|
|
events = [];
|
|
|
|
const mock = createMockModel({
|
|
id: "gpt-5.5",
|
|
provider: "openai",
|
|
contextWindow: 10_000,
|
|
responses: [
|
|
{
|
|
content: ["seed response"],
|
|
stopReason: "stop",
|
|
usage: {
|
|
input: 8_500,
|
|
output: 10,
|
|
cacheRead: 0,
|
|
cacheWrite: 0,
|
|
totalTokens: 8_510,
|
|
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
|
|
},
|
|
},
|
|
{ content: ["ok"], stopReason: "stop" },
|
|
],
|
|
});
|
|
const agent = new Agent({
|
|
getApiKey: () => "test-key",
|
|
initialState: {
|
|
model: mock,
|
|
systemPrompt: ["expanded system prompt ".repeat(30_000)],
|
|
tools: [],
|
|
messages: [],
|
|
},
|
|
streamFn: mock.stream,
|
|
});
|
|
session = new AgentSession({
|
|
agent,
|
|
sessionManager,
|
|
settings: Settings.isolated({
|
|
"compaction.enabled": false,
|
|
"compaction.autoContinue": false,
|
|
"compaction.strategy": "context-full",
|
|
"compaction.thresholdTokens": 9_500,
|
|
"contextPromotion.enabled": false,
|
|
}),
|
|
modelRegistry,
|
|
});
|
|
session.subscribe(event => {
|
|
events.push(event);
|
|
});
|
|
|
|
await session.prompt("seed prompt");
|
|
expect(mock.calls).toHaveLength(1);
|
|
session.settings.set("compaction.enabled", true);
|
|
const compactSpy = vi.spyOn(compactionModule, "compact").mockImplementation(async preparation => ({
|
|
summary: "pre-prompt compacted",
|
|
shortSummary: undefined,
|
|
firstKeptEntryId: preparation.firstKeptEntryId,
|
|
tokensBefore: preparation.tokensBefore,
|
|
details: {},
|
|
}));
|
|
|
|
await session.prompt("small pending prompt");
|
|
await drainMaintenance();
|
|
|
|
expect(compactSpy).not.toHaveBeenCalled();
|
|
expect(events.filter(event => event.type === "auto_compaction_start")).toHaveLength(0);
|
|
expect(mock.calls).toHaveLength(2);
|
|
});
|
|
it("does not run auto maintenance after final yield", async () => {
|
|
session.settings.set("compaction.strategy", "handoff");
|
|
session.settings.set("compaction.thresholdPercent", 1);
|
|
session.settings.set("contextPromotion.enabled", false);
|
|
|
|
const model = session.model;
|
|
if (!model) {
|
|
throw new Error("Expected model to be set");
|
|
}
|
|
|
|
const yieldCall: ToolCall = {
|
|
type: "toolCall",
|
|
id: "call_yield_done",
|
|
name: "yield",
|
|
arguments: { result: { data: { done: true } } },
|
|
};
|
|
const assistantMessage: AssistantMessage = {
|
|
role: "assistant",
|
|
content: [yieldCall],
|
|
api: model.api,
|
|
provider: model.provider,
|
|
model: model.id,
|
|
stopReason: "toolUse",
|
|
usage: {
|
|
input: 10_000,
|
|
output: 1_000,
|
|
cacheRead: 0,
|
|
cacheWrite: 0,
|
|
totalTokens: 11_000,
|
|
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
|
|
},
|
|
timestamp: Date.now(),
|
|
};
|
|
|
|
const handoffSpy = vi.spyOn(session, "handoff").mockResolvedValue({ document: "handoff document" });
|
|
session.agent.emitExternalEvent({ type: "message_end", message: assistantMessage });
|
|
session.agent.emitExternalEvent({
|
|
type: "tool_execution_end",
|
|
toolCallId: yieldCall.id,
|
|
toolName: "yield",
|
|
result: {
|
|
content: [{ type: "text", text: "Result submitted." }],
|
|
details: { status: "success", data: { done: true } },
|
|
},
|
|
isError: false,
|
|
});
|
|
session.agent.emitExternalEvent({ type: "agent_end", messages: [assistantMessage] });
|
|
await drainMaintenance();
|
|
|
|
expect(handoffSpy).not.toHaveBeenCalled();
|
|
expect(events.filter(event => event.type === "auto_compaction_start")).toHaveLength(0);
|
|
expect(events.filter(event => event.type === "auto_compaction_end")).toHaveLength(0);
|
|
});
|
|
|
|
it("persists handoff session immediately with previous session as parent", async () => {
|
|
const previousSessionFile = session.sessionFile;
|
|
if (!previousSessionFile) {
|
|
throw new Error("Expected previous session file");
|
|
}
|
|
|
|
const handoffText = "## Goal\nContinue from here";
|
|
vi.spyOn(compactionModule, "generateHandoff").mockResolvedValue(handoffText);
|
|
|
|
const result = await session.handoff();
|
|
const handoffSessionFile = session.sessionFile;
|
|
if (!handoffSessionFile) {
|
|
throw new Error("Expected handoff session file");
|
|
}
|
|
|
|
type PersistedEntry = {
|
|
type?: string;
|
|
parentSession?: string;
|
|
customType?: string;
|
|
display?: boolean;
|
|
};
|
|
const handoffEntries = (await Bun.file(handoffSessionFile).text())
|
|
.trim()
|
|
.split("\n")
|
|
.map(line => JSON.parse(line) as PersistedEntry);
|
|
|
|
expect(result?.document).toBe(handoffText);
|
|
expect(session.getLastAssistantText()).toBeUndefined();
|
|
expect(session.hasCopyCandidateAssistantMessage()).toBe(false);
|
|
expect(session.getLastVisibleHandoffText()).toBe(
|
|
`<handoff-context>\n${handoffText}\n</handoff-context>\n\nThe above is a handoff document from a previous session. Use this context to continue the work seamlessly.`,
|
|
);
|
|
expect(handoffSessionFile).not.toBe(previousSessionFile);
|
|
expect(handoffEntries[0]).toMatchObject({ type: "session", parentSession: previousSessionFile });
|
|
expect(
|
|
handoffEntries.some(
|
|
entry => entry.type === "custom_message" && entry.customType === "handoff" && entry.display,
|
|
),
|
|
).toBe(true);
|
|
|
|
const previousSessionText = await Bun.file(previousSessionFile).text();
|
|
expect(previousSessionText).toContain('"text":"seed"');
|
|
});
|
|
|
|
it("does not run auto maintenance when strategy is off", async () => {
|
|
session.settings.set("compaction.strategy", "off");
|
|
session.settings.set("compaction.thresholdPercent", 1);
|
|
session.settings.set("contextPromotion.enabled", false);
|
|
|
|
const model = session.model;
|
|
if (!model) {
|
|
throw new Error("Expected model to be set");
|
|
}
|
|
|
|
const assistantMessage: AssistantMessage = {
|
|
role: "assistant",
|
|
content: [{ type: "text", text: "maintenance trigger" }],
|
|
api: model.api,
|
|
provider: model.provider,
|
|
model: model.id,
|
|
stopReason: "stop",
|
|
usage: {
|
|
input: 10_000,
|
|
output: 1_000,
|
|
cacheRead: 0,
|
|
cacheWrite: 0,
|
|
totalTokens: 11_000,
|
|
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
|
|
},
|
|
timestamp: Date.now(),
|
|
};
|
|
|
|
const handoffSpy = vi.spyOn(session, "handoff");
|
|
session.agent.emitExternalEvent({ type: "message_end", message: assistantMessage });
|
|
session.agent.emitExternalEvent({ type: "agent_end", messages: [assistantMessage] });
|
|
await drainMaintenance();
|
|
|
|
expect(handoffSpy).not.toHaveBeenCalled();
|
|
expect(events.filter(event => event.type === "auto_compaction_start")).toHaveLength(0);
|
|
expect(events.filter(event => event.type === "auto_compaction_end")).toHaveLength(0);
|
|
});
|
|
|
|
it("restores context-full strategy when enabling auto-compaction from off strategy", () => {
|
|
session.settings.set("compaction.enabled", true);
|
|
session.settings.set("compaction.strategy", "off");
|
|
|
|
expect(session.autoCompactionEnabled).toBe(false);
|
|
session.setAutoCompactionEnabled(true);
|
|
expect(session.settings.get("compaction.strategy")).toBe("context-full");
|
|
expect(session.autoCompactionEnabled).toBe(true);
|
|
});
|
|
|
|
it("falls back to context-full maintenance for overflow when strategy is handoff", async () => {
|
|
session.settings.set("compaction.strategy", "handoff");
|
|
session.settings.set("contextPromotion.enabled", false);
|
|
|
|
const model = session.model;
|
|
if (!model) {
|
|
throw new Error("Expected model to be set");
|
|
}
|
|
const handoffSpy = vi.spyOn(session, "handoff");
|
|
|
|
const overflowAssistant: AssistantMessage = {
|
|
role: "assistant",
|
|
content: [{ type: "text", text: "overflow" }],
|
|
api: model.api,
|
|
provider: model.provider,
|
|
model: model.id,
|
|
stopReason: "error",
|
|
errorMessage: "maximum context length is 200000 tokens, however you requested 200001 tokens",
|
|
usage: {
|
|
input: 120_000,
|
|
output: 0,
|
|
cacheRead: 0,
|
|
cacheWrite: 0,
|
|
totalTokens: 120_000,
|
|
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
|
|
},
|
|
timestamp: Date.now(),
|
|
};
|
|
|
|
session.agent.emitExternalEvent({ type: "message_end", message: overflowAssistant });
|
|
session.agent.emitExternalEvent({ type: "agent_end", messages: [overflowAssistant] });
|
|
await waitFor(() => events.filter(event => event.type === "auto_compaction_end").length === 1);
|
|
|
|
expect(handoffSpy).not.toHaveBeenCalled();
|
|
const startEvents = events.filter(event => event.type === "auto_compaction_start");
|
|
expect(startEvents).toHaveLength(1);
|
|
expect(startEvents[0]).toMatchObject({ type: "auto_compaction_start", reason: "overflow" });
|
|
const endEvents = events.filter(event => event.type === "auto_compaction_end");
|
|
expect(endEvents).toHaveLength(1);
|
|
expect(endEvents[0]).not.toMatchObject({
|
|
errorMessage: "Auto-handoff failed: no handoff document was generated",
|
|
});
|
|
});
|
|
|
|
it("uses handoff strategy for threshold-triggered auto maintenance", async () => {
|
|
session.settings.set("compaction.strategy", "handoff");
|
|
session.settings.set("compaction.thresholdPercent", 1);
|
|
session.settings.set("contextPromotion.enabled", false);
|
|
|
|
const model = session.model;
|
|
if (!model) {
|
|
throw new Error("Expected model to be set");
|
|
}
|
|
|
|
const assistantMessage: AssistantMessage = {
|
|
role: "assistant",
|
|
content: [{ type: "text", text: "maintenance trigger" }],
|
|
api: model.api,
|
|
provider: model.provider,
|
|
model: model.id,
|
|
stopReason: "stop",
|
|
usage: {
|
|
input: 10_000,
|
|
output: 1_000,
|
|
cacheRead: 0,
|
|
cacheWrite: 0,
|
|
totalTokens: 11_000,
|
|
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
|
|
},
|
|
timestamp: Date.now(),
|
|
};
|
|
|
|
const handoffSpy = vi.spyOn(session, "handoff").mockResolvedValue({ document: "handoff document" });
|
|
|
|
session.agent.emitExternalEvent({ type: "message_end", message: assistantMessage });
|
|
session.agent.emitExternalEvent({ type: "agent_end", messages: [assistantMessage] });
|
|
await waitFor(
|
|
() =>
|
|
handoffSpy.mock.calls.length === 1 &&
|
|
events.filter(event => event.type === "auto_compaction_end").length === 1,
|
|
);
|
|
|
|
expect(handoffSpy).toHaveBeenCalledTimes(1);
|
|
expect(handoffSpy).toHaveBeenCalledWith(expect.stringContaining("Threshold-triggered maintenance"), {
|
|
autoTriggered: true,
|
|
signal: expect.anything(),
|
|
});
|
|
expect(events.filter(event => event.type === "auto_compaction_start")).toHaveLength(1);
|
|
const endEvents = events.filter(event => event.type === "auto_compaction_end");
|
|
expect(endEvents).toHaveLength(1);
|
|
expect(endEvents[0]).toMatchObject({ type: "auto_compaction_end", aborted: false, willRetry: false });
|
|
});
|
|
|
|
it("completes threshold-triggered auto-handoff while the original prompt is still unwinding", async () => {
|
|
authStorage.setRuntimeApiKey("anthropic", "test-key");
|
|
const model = getBundledModel("anthropic", "claude-sonnet-4-5");
|
|
if (!model) {
|
|
throw new Error("Expected built-in anthropic model to exist");
|
|
}
|
|
|
|
await session.dispose();
|
|
sessionManager = SessionManager.create(tempDir.path(), tempDir.path());
|
|
events = [];
|
|
sessionManager.appendMessage({
|
|
role: "user",
|
|
content: [{ type: "text", text: "seed" }],
|
|
timestamp: Date.now() - 2,
|
|
});
|
|
sessionManager.appendMessage({
|
|
role: "assistant",
|
|
content: [{ type: "text", text: "seed response" }],
|
|
api: model.api,
|
|
provider: model.provider,
|
|
model: model.id,
|
|
stopReason: "stop",
|
|
usage: {
|
|
input: 16,
|
|
output: 8,
|
|
cacheRead: 0,
|
|
cacheWrite: 0,
|
|
totalTokens: 24,
|
|
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
|
|
},
|
|
timestamp: Date.now() - 1,
|
|
});
|
|
|
|
const mock = createMockModel({
|
|
responses: [
|
|
{
|
|
content: [{ type: "text", text: "maintenance trigger" }],
|
|
stopReason: "stop",
|
|
usage: {
|
|
input: 190_000,
|
|
output: 1_000,
|
|
cacheRead: 0,
|
|
cacheWrite: 0,
|
|
totalTokens: 191_000,
|
|
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
|
|
},
|
|
},
|
|
],
|
|
});
|
|
|
|
const agent = new Agent({
|
|
getApiKey: () => "test-key",
|
|
initialState: {
|
|
model,
|
|
systemPrompt: ["Test"],
|
|
tools: [],
|
|
messages: [],
|
|
},
|
|
streamFn: mock.stream,
|
|
});
|
|
|
|
session = new AgentSession({
|
|
agent,
|
|
sessionManager,
|
|
settings: Settings.isolated({
|
|
"compaction.enabled": true,
|
|
"compaction.autoContinue": false,
|
|
"compaction.strategy": "handoff",
|
|
"compaction.thresholdPercent": 1,
|
|
"contextPromotion.enabled": false,
|
|
}),
|
|
modelRegistry,
|
|
});
|
|
session.subscribe(event => {
|
|
events.push(event);
|
|
});
|
|
|
|
const generateHandoffSpy = vi
|
|
.spyOn(compactionModule, "generateHandoff")
|
|
.mockResolvedValue("## Goal\nContinue from here");
|
|
await session.prompt("Trigger threshold handoff");
|
|
|
|
expect(mock.calls).toHaveLength(1);
|
|
expect(generateHandoffSpy).toHaveBeenCalledTimes(1);
|
|
const endEvents = events.filter(event => event.type === "auto_compaction_end");
|
|
expect(endEvents).toHaveLength(1);
|
|
expect(endEvents[0]).toMatchObject({ type: "auto_compaction_end", action: "handoff", aborted: false });
|
|
expect(endEvents[0]).not.toMatchObject({ errorMessage: expect.any(String) });
|
|
expect(sessionManager.getEntries().filter(entry => entry.type === "compaction")).toHaveLength(0);
|
|
});
|
|
|
|
it("does not start agent.continue when threshold-handoff defers and todos are incomplete", async () => {
|
|
// Reproduces the user-reported race: at agent_end, threshold + handoff strategy
|
|
// schedules a deferred handoff and returns. The handler used to fall through to
|
|
// #checkTodoCompletion, which scheduled agent.continue() — both fired concurrently,
|
|
// rendering as "Auto-handoff" loader + an assistant message still streaming.
|
|
session.settings.set("compaction.strategy", "handoff");
|
|
session.settings.set("compaction.thresholdPercent", 1);
|
|
session.settings.set("contextPromotion.enabled", false);
|
|
session.settings.set("todo.enabled", true);
|
|
session.settings.set("todo.reminders", true);
|
|
|
|
// Active todo phase with an incomplete task so #checkTodoCompletion would normally fire.
|
|
session.setTodoPhases([{ name: "Phase 1", tasks: [{ content: "unfinished work", status: "pending" }] }]);
|
|
|
|
const model = session.model;
|
|
if (!model) {
|
|
throw new Error("Expected model to be set");
|
|
}
|
|
|
|
const handoffSpy = vi
|
|
.spyOn(session, "handoff")
|
|
.mockResolvedValue({ document: "## Goal\nContinue", savedPath: undefined });
|
|
const continueSpy = vi.spyOn(session.agent, "continue");
|
|
|
|
const assistantMessage: AssistantMessage = {
|
|
role: "assistant",
|
|
content: [{ type: "text", text: "maintenance trigger" }],
|
|
api: model.api,
|
|
provider: model.provider,
|
|
model: model.id,
|
|
stopReason: "stop",
|
|
usage: {
|
|
input: 10_000,
|
|
output: 1_000,
|
|
cacheRead: 0,
|
|
cacheWrite: 0,
|
|
totalTokens: 11_000,
|
|
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
|
|
},
|
|
timestamp: Date.now(),
|
|
};
|
|
|
|
session.agent.emitExternalEvent({ type: "message_end", message: assistantMessage });
|
|
session.agent.emitExternalEvent({ type: "agent_end", messages: [assistantMessage] });
|
|
await waitFor(() => handoffSpy.mock.calls.length === 1);
|
|
await session.waitForIdle();
|
|
|
|
expect(handoffSpy).toHaveBeenCalledTimes(1);
|
|
// The bug surfaced as agent.continue() racing the deferred handoff. With the fix,
|
|
// the agent_end handler short-circuits after the deferred-handoff signal.
|
|
expect(continueSpy).not.toHaveBeenCalled();
|
|
});
|
|
|
|
it("dispose unblocks the post-prompt drain when a deferred handoff is mid-flight", async () => {
|
|
// Reproduces /exit / Ctrl+C-double-tap hanging when a deferred handoff is awaiting
|
|
// the LLM call: dispose() now aborts the handoff controller before draining post-prompt
|
|
// tasks, so Promise.allSettled() in #cancelPostPromptTasks can resolve.
|
|
session.settings.set("compaction.strategy", "handoff");
|
|
session.settings.set("compaction.thresholdPercent", 1);
|
|
session.settings.set("contextPromotion.enabled", false);
|
|
|
|
const model = session.model;
|
|
if (!model) {
|
|
throw new Error("Expected model to be set");
|
|
}
|
|
|
|
const { promise: handoffPending, resolve: resolveHandoff } = Promise.withResolvers<string>();
|
|
|
|
const generateHandoffSpy = vi
|
|
.spyOn(compactionModule, "generateHandoff")
|
|
.mockImplementation(async (_msgs, _model, _key, _opts, signal) => {
|
|
// Mirror the real generateHandoff contract: reject when the caller aborts.
|
|
return await new Promise<string>((resolve, reject) => {
|
|
signal?.addEventListener("abort", () => reject(new Error("Handoff cancelled")), { once: true });
|
|
handoffPending.then(resolve, reject);
|
|
});
|
|
});
|
|
|
|
const assistantMessage: AssistantMessage = {
|
|
role: "assistant",
|
|
content: [{ type: "text", text: "maintenance trigger" }],
|
|
api: model.api,
|
|
provider: model.provider,
|
|
model: model.id,
|
|
stopReason: "stop",
|
|
usage: {
|
|
input: 10_000,
|
|
output: 1_000,
|
|
cacheRead: 0,
|
|
cacheWrite: 0,
|
|
totalTokens: 11_000,
|
|
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
|
|
},
|
|
timestamp: Date.now(),
|
|
};
|
|
|
|
session.agent.emitExternalEvent({ type: "message_end", message: assistantMessage });
|
|
session.agent.emitExternalEvent({ type: "agent_end", messages: [assistantMessage] });
|
|
// Let the deferred handoff post-prompt task enter the generateHandoff await.
|
|
await waitFor(() => session.isGeneratingHandoff);
|
|
expect(generateHandoffSpy).toHaveBeenCalledTimes(1);
|
|
expect(session.isGeneratingHandoff).toBe(true);
|
|
|
|
// dispose must NOT wait for the LLM call to resolve on its own — it must abort it.
|
|
const disposed = Promise.race([
|
|
session.dispose().then(() => "disposed" as const),
|
|
Bun.sleep(2_000).then(() => "timeout" as const),
|
|
]);
|
|
|
|
await expect(disposed).resolves.toBe("disposed");
|
|
// Releasing after the fact must not leak into other tests.
|
|
resolveHandoff("handoff");
|
|
});
|
|
|
|
it("falls back to context-full when handoff strategy returns no document", async () => {
|
|
session.settings.set("compaction.strategy", "handoff");
|
|
session.settings.set("compaction.thresholdPercent", 1);
|
|
session.settings.set("contextPromotion.enabled", false);
|
|
|
|
const model = session.model;
|
|
if (!model) {
|
|
throw new Error("Expected model to be set");
|
|
}
|
|
|
|
const assistantMessage: AssistantMessage = {
|
|
role: "assistant",
|
|
content: [{ type: "text", text: "maintenance trigger" }],
|
|
api: model.api,
|
|
provider: model.provider,
|
|
model: model.id,
|
|
stopReason: "stop",
|
|
usage: {
|
|
input: 10_000,
|
|
output: 1_000,
|
|
cacheRead: 0,
|
|
cacheWrite: 0,
|
|
totalTokens: 11_000,
|
|
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
|
|
},
|
|
timestamp: Date.now(),
|
|
};
|
|
|
|
const handoffSpy = vi.spyOn(session, "handoff").mockResolvedValue(undefined);
|
|
|
|
session.agent.emitExternalEvent({ type: "message_end", message: assistantMessage });
|
|
session.agent.emitExternalEvent({ type: "agent_end", messages: [assistantMessage] });
|
|
await waitFor(() => events.filter(event => event.type === "auto_compaction_end").length === 1);
|
|
|
|
expect(handoffSpy).toHaveBeenCalledTimes(1);
|
|
const endEvents = events.filter(event => event.type === "auto_compaction_end");
|
|
expect(endEvents).toHaveLength(1);
|
|
expect(endEvents[0]).toMatchObject({
|
|
type: "auto_compaction_end",
|
|
action: "context-full",
|
|
aborted: false,
|
|
willRetry: false,
|
|
});
|
|
expect(endEvents[0]).not.toMatchObject({
|
|
errorMessage: "Auto-handoff failed: no handoff document was generated",
|
|
});
|
|
});
|
|
|
|
it("resets to the base system prompt before generating a handoff", async () => {
|
|
const model = session.model;
|
|
if (!model) {
|
|
throw new Error("Expected model to be set");
|
|
}
|
|
await session.dispose();
|
|
sessionManager = SessionManager.create(tempDir.path(), tempDir.path());
|
|
|
|
const extensionsResult = await loadExtensions([], tempDir.path());
|
|
const extensionRunner = new ExtensionRunner(
|
|
extensionsResult.extensions,
|
|
extensionsResult.runtime,
|
|
tempDir.path(),
|
|
sessionManager,
|
|
modelRegistry,
|
|
);
|
|
const emitBeforeAgentStart = vi.spyOn(extensionRunner, "emitBeforeAgentStart").mockResolvedValueOnce({
|
|
systemPrompt: ["Hook override"],
|
|
});
|
|
vi.spyOn(extensionRunner, "emit").mockResolvedValue(undefined);
|
|
|
|
const mock = createMockModel({
|
|
responses: [{ content: ["normal response"] }],
|
|
});
|
|
const agent = new Agent({
|
|
getApiKey: () => "test-key",
|
|
initialState: {
|
|
model,
|
|
systemPrompt: ["Test"],
|
|
tools: [],
|
|
messages: [],
|
|
},
|
|
streamFn: mock.stream,
|
|
});
|
|
|
|
session = new AgentSession({
|
|
agent,
|
|
sessionManager,
|
|
settings: Settings.isolated({ "compaction.enabled": false }),
|
|
modelRegistry,
|
|
extensionRunner,
|
|
});
|
|
sessionManager.appendMessage({
|
|
role: "user",
|
|
content: [{ type: "text", text: "seed" }],
|
|
timestamp: Date.now() - 2,
|
|
});
|
|
sessionManager.appendMessage({
|
|
role: "assistant",
|
|
content: [{ type: "text", text: "seed response" }],
|
|
api: model.api,
|
|
provider: model.provider,
|
|
model: model.id,
|
|
stopReason: "stop",
|
|
usage: {
|
|
input: 16,
|
|
output: 8,
|
|
cacheRead: 0,
|
|
cacheWrite: 0,
|
|
totalTokens: 24,
|
|
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
|
|
},
|
|
timestamp: Date.now() - 1,
|
|
});
|
|
|
|
await session.prompt("hello from user");
|
|
const generateHandoffSpy = vi
|
|
.spyOn(compactionModule, "generateHandoff")
|
|
.mockResolvedValue("## Goal\nContinue from here");
|
|
await session.handoff();
|
|
|
|
expect(emitBeforeAgentStart).toHaveBeenCalledTimes(1);
|
|
expect(mock.calls.map(c => c.context.systemPrompt?.join("\n\n") ?? "")).toEqual(["Hook override"]);
|
|
const handoffCall = generateHandoffSpy.mock.calls[0];
|
|
if (!handoffCall) throw new Error("Expected generateHandoff call");
|
|
expect(handoffCall[3].systemPrompt).toEqual(["Test"]);
|
|
});
|
|
|
|
it("saves auto-handoff document to disk when enabled", async () => {
|
|
session.settings.set("compaction.handoffSaveToDisk", true);
|
|
|
|
const handoffText = "## Goal\nContinue from here";
|
|
vi.spyOn(compactionModule, "generateHandoff").mockResolvedValue(handoffText);
|
|
|
|
const result = await session.handoff(undefined, { autoTriggered: true });
|
|
expect(result?.savedPath).toBeDefined();
|
|
if (!result?.savedPath) throw new Error("Expected handoff document path");
|
|
expect(result.savedPath.endsWith(".md")).toBe(true);
|
|
const savedText = await Bun.file(result.savedPath).text();
|
|
expect(savedText).toContain(handoffText);
|
|
});
|
|
|
|
it("does not save manual handoff document when save setting is enabled", async () => {
|
|
session.settings.set("compaction.handoffSaveToDisk", true);
|
|
|
|
vi.spyOn(compactionModule, "generateHandoff").mockResolvedValue("## Goal\nManual handoff");
|
|
|
|
const result = await session.handoff();
|
|
expect(result?.savedPath).toBeUndefined();
|
|
});
|
|
|
|
it("does not start handoff prompt when provided signal is already cancelled", async () => {
|
|
const controller = new AbortController();
|
|
controller.abort();
|
|
|
|
const generateHandoffSpy = vi.spyOn(compactionModule, "generateHandoff");
|
|
|
|
await expect(session.handoff(undefined, { signal: controller.signal })).rejects.toThrow("Handoff cancelled");
|
|
expect(generateHandoffSpy).not.toHaveBeenCalled();
|
|
});
|
|
|
|
it("aborts handoff generation when provided signal is cancelled", async () => {
|
|
const controller = new AbortController();
|
|
const started = Promise.withResolvers<void>();
|
|
const cancelled = Promise.withResolvers<string>();
|
|
const generateHandoffSpy = vi
|
|
.spyOn(compactionModule, "generateHandoff")
|
|
.mockImplementation((_messages, _model, _apiKey, _options, signal) => {
|
|
started.resolve();
|
|
const onAbort = () => {
|
|
const error = new Error("aborted");
|
|
error.name = "AbortError";
|
|
cancelled.reject(error);
|
|
};
|
|
if (signal?.aborted) {
|
|
onAbort();
|
|
} else {
|
|
signal?.addEventListener("abort", onAbort, { once: true });
|
|
}
|
|
return cancelled.promise;
|
|
});
|
|
|
|
const handoffPromise = session.handoff(undefined, { signal: controller.signal });
|
|
await started.promise;
|
|
controller.abort();
|
|
|
|
await expect(handoffPromise).rejects.toThrow("Handoff cancelled");
|
|
expect(generateHandoffSpy).toHaveBeenCalledTimes(1);
|
|
expect(generateHandoffSpy.mock.calls[0]?.[4]?.aborted).toBe(true);
|
|
});
|
|
});
|