fix(coding-agent): avoid pre-prompt compaction continue

packages/coding-agent/src/core/agent-session.ts

@@ -1075,17 +1075,11 @@ export class AgentSession {
 				throw new Error(formatNoApiKeyFoundMessage(this.model.provider));
 			}
 
-			// Check if we need to compact before sending (catches aborted responses)
+			// Check if we need to compact before sending (catches aborted responses).
+			// The user's new prompt is sent below, so do not call agent.continue() here.
 			const lastAssistant = this._findLastAssistantMessage();
-			if (lastAssistant && (await this._checkCompaction(lastAssistant, false))) {
-				try {
-					await this.agent.continue();
-					while (await this._handlePostAgentRun()) {
-						await this.agent.continue();
-					}
-				} finally {
-					this._flushPendingBashMessages();
-				}
+			if (lastAssistant) {
+				await this._checkCompaction(lastAssistant, false);
 			}
 
 			// Build messages array (custom message if any, then user message)

packages/coding-agent/test/suite/regressions/pre-prompt-compaction-no-continue.test.ts

@@ -0,0 +1,75 @@
+import { type AssistantMessage, fauxAssistantMessage } from "@earendil-works/pi-ai";
+import { afterEach, describe, expect, it, vi } from "vitest";
+import { createHarness, getUserTexts, type Harness } from "../harness.ts";
+
+function createUsage(totalTokens: number) {
+	return {
+		input: totalTokens,
+		output: 0,
+		cacheRead: 0,
+		cacheWrite: 0,
+		totalTokens,
+		cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
+	};
+}
+
+describe("pre-prompt compaction regression", () => {
+	const harnesses: Harness[] = [];
+
+	afterEach(() => {
+		vi.restoreAllMocks();
+		while (harnesses.length > 0) {
+			harnesses.pop()?.cleanup();
+		}
+	});
+
+	it("compacts length-stop overflow before a new prompt without continuing from an assistant message", async () => {
+		const harness = await createHarness({
+			models: [{ id: "faux-1", contextWindow: 100, maxTokens: 100 }],
+			settings: { compaction: { enabled: true, keepRecentTokens: 1, reserveTokens: 0 } },
+			extensionFactories: [
+				(pi) => {
+					pi.on("session_before_compact", async (event) => ({
+						compaction: {
+							summary: "pre-prompt summary",
+							firstKeptEntryId: event.preparation.firstKeptEntryId,
+							tokensBefore: event.preparation.tokensBefore,
+							details: {},
+						},
+					}));
+				},
+			],
+		});
+		harnesses.push(harness);
+
+		const now = Date.now();
+		const model = harness.getModel();
+		harness.sessionManager.appendMessage({
+			role: "user",
+			content: [{ type: "text", text: "previous prompt" }],
+			timestamp: now - 1000,
+		});
+		const lengthStopAssistant: AssistantMessage = {
+			...fauxAssistantMessage("length-stop assistant response", { stopReason: "length", timestamp: now - 500 }),
+			api: model.api,
+			provider: model.provider,
+			model: model.id,
+			usage: createUsage(100),
+		};
+		harness.sessionManager.appendMessage(lengthStopAssistant);
+		harness.session.agent.state.messages = harness.sessionManager.buildSessionContext().messages;
+		harness.setResponses([fauxAssistantMessage("answered next prompt")]);
+		const continueSpy = vi.spyOn(harness.session.agent, "continue");
+
+		await expect(harness.session.prompt("next prompt")).resolves.toBeUndefined();
+
+		expect(continueSpy).not.toHaveBeenCalled();
+		expect(harness.eventsOfType("compaction_end").at(-1)).toMatchObject({
+			reason: "overflow",
+			aborted: false,
+			willRetry: true,
+		});
+		expect(getUserTexts(harness)).toContain("next prompt");
+		expect(harness.faux.state.callCount).toBe(1);
+	});
+});