diff --git a/packages/coding-agent/test/agent-session-eager-compaction.test.ts b/packages/coding-agent/test/agent-session-eager-compaction.test.ts
index 058926fe1..a7ad2b7e4 100644
--- a/packages/coding-agent/test/agent-session-eager-compaction.test.ts
+++ b/packages/coding-agent/test/agent-session-eager-compaction.test.ts
@@ -21,9 +21,10 @@ import { TempDir } from "@oh-my-pi/pi-utils";
 // the delegate-via-tasks / phased-todo guidance. The post-compaction auto-continuation
 // turn must carry the gated reminders again (reminder-only — never a forced tool_choice).
 
-const CONTINUE_MARKER = "Resume work on the user's most recent intent";
+const TASK_DELEGATION_MARKER = "Task delegation enabled";
 
 type ObservedPromptCall = {
+	callIndex: number;
 	toolChoice: string | undefined;
 	messageTexts: string[];
 };
@@ -202,6 +203,7 @@ describe("AgentSession eager prelude re-injection after compaction", () => {
 			getToolChoice: () => session?.nextToolChoiceDirective(),
 			streamFn: (_model, context, options) => {
 				const call: ObservedPromptCall = {
+					callIndex: observedCalls.length,
 					toolChoice: getToolChoiceName(options?.toolChoice),
 					messageTexts: context.messages.map(message => getMessageText(message)),
 				};
@@ -276,7 +278,7 @@ describe("AgentSession eager prelude re-injection after compaction", () => {
 		activateOngoingGoal(session);
 		await session.prompt("refactor the parser across modules");
 		emitHighUsageTurn(session);
-		return waitForCall(call => call.messageTexts.some(text => text.includes(CONTINUE_MARKER)));
+		return waitForCall(call => call.callIndex > 0);
 	}
 
 	it("re-injects the eager task reminder on the auto-continuation turn (task.eager always)", async () => {
@@ -285,7 +287,7 @@ describe("AgentSession eager prelude re-injection after compaction", () => {
 
 		const continuation = await runToContinuation(session, waitForCall);
 
-		const reminder = continuation.messageTexts.find(text => text.includes("delegation is enabled"));
+		const reminder = continuation.messageTexts.find(text => text.includes(TASK_DELEGATION_MARKER));
 		expect(reminder).toBeDefined();
 		expect(reminder).toContain("`task`");
 		// Reminder-only: the post-compaction nudge never forces a tool on the resumed turn.
@@ -298,7 +300,7 @@ describe("AgentSession eager prelude re-injection after compaction", () => {
 
 		const continuation = await runToContinuation(session, waitForCall);
 
-		expect(continuation.messageTexts.some(text => text.includes("delegation is enabled"))).toBe(false);
+		expect(continuation.messageTexts.some(text => text.includes(TASK_DELEGATION_MARKER))).toBe(false);
 	});
 
 	it("does not re-inject the eager task reminder when task.eager is preferred", async () => {
@@ -307,7 +309,7 @@ describe("AgentSession eager prelude re-injection after compaction", () => {
 
 		const continuation = await runToContinuation(session, waitForCall);
 
-		expect(continuation.messageTexts.some(text => text.includes("delegation is enabled"))).toBe(false);
+		expect(continuation.messageTexts.some(text => text.includes(TASK_DELEGATION_MARKER))).toBe(false);
 	});
 
 	it("does not re-inject the eager task reminder for subagent sessions", async () => {
@@ -316,7 +318,7 @@ describe("AgentSession eager prelude re-injection after compaction", () => {
 
 		const continuation = await runToContinuation(session, waitForCall);
 
-		expect(continuation.messageTexts.some(text => text.includes("delegation is enabled"))).toBe(false);
+		expect(continuation.messageTexts.some(text => text.includes(TASK_DELEGATION_MARKER))).toBe(false);
 	});
 
 	it("does not re-inject the eager task reminder in plan mode", async () => {
@@ -326,7 +328,7 @@ describe("AgentSession eager prelude re-injection after compaction", () => {
 
 		const continuation = await runToContinuation(session, waitForCall);
 
-		expect(continuation.messageTexts.some(text => text.includes("delegation is enabled"))).toBe(false);
+		expect(continuation.messageTexts.some(text => text.includes(TASK_DELEGATION_MARKER))).toBe(false);
 	});
 
 	it("re-injects the eager todo reminder on the auto-continuation turn (todo.eager preferred)", async () => {
@@ -373,7 +375,7 @@ describe("AgentSession eager prelude re-injection after compaction", () => {
 		});
 		stubCompaction(todoEntryId);
 
-		const continuationPromise = waitForCall(call => call.messageTexts.some(text => text.includes(CONTINUE_MARKER)));
+		const continuationPromise = waitForCall(call => call.callIndex > 0);
 		emitHighUsageTurn(session);
 		const continuation = await continuationPromise;
 
diff --git a/packages/coding-agent/test/agent-session-plan-reference-compaction.test.ts b/packages/coding-agent/test/agent-session-plan-reference-compaction.test.ts
index 5524ae7c7..749148b07 100644
--- a/packages/coding-agent/test/agent-session-plan-reference-compaction.test.ts
+++ b/packages/coding-agent/test/agent-session-plan-reference-compaction.test.ts
@@ -30,9 +30,7 @@ import { AuthStorage } from "../src/session/auth-storage";
 import { convertToLlm } from "../src/session/messages";
 import { SessionManager } from "../src/session/session-manager";
 
-const CONTINUE_MARKER = "Resume work on the user's most recent intent";
-
-type ObservedPromptCall = { messageTexts: string[] };
+type ObservedPromptCall = { callIndex: number; messageTexts: string[] };
 
 type Harness = {
 	session: AgentSession;
@@ -163,8 +161,11 @@ describe("AgentSession approved-plan reference re-injection after compaction (is
 			convertToLlm,
 			getToolChoice: () => session?.nextToolChoiceDirective(),
 			streamFn: (_model, context) => {
-				observedCalls.push({ messageTexts: context.messages.map(message => getMessageText(message)) });
-				const call = observedCalls[observedCalls.length - 1];
+				const call = {
+					callIndex: observedCalls.length,
+					messageTexts: context.messages.map(message => getMessageText(message)),
+				};
+				observedCalls.push(call);
 				for (let i = waiters.length - 1; i >= 0; i--) {
 					const waiter = waiters[i];
 					if (waiter?.predicate(call)) {
@@ -256,7 +257,7 @@ describe("AgentSession approved-plan reference re-injection after compaction (is
 		// Auto-compaction fires, replacing history (dropping the delivered reference),
 		// then schedules the auto-continuation turn.
 		emitHighUsageTurn(session);
-		const continuation = await waitForCall(call => call.messageTexts.some(text => text.includes(CONTINUE_MARKER)));
+		const continuation = await waitForCall(call => call.callIndex > 0);
 
 		// The post-compaction continuation MUST carry the durable plan reference again.
 		expect(continuation.messageTexts.some(text => text.includes(planMarker))).toBe(false);
@@ -280,7 +281,7 @@ describe("AgentSession approved-plan reference re-injection after compaction (is
 		expect(firstCall.messageTexts.some(text => text.includes(planMarker))).toBe(false);
 
 		emitHighUsageTurn(session);
-		const continuation = await waitForCall(call => call.messageTexts.some(text => text.includes(CONTINUE_MARKER)));
+		const continuation = await waitForCall(call => call.callIndex > 0);
 
 		expect(continuation.messageTexts.some(text => text.includes(planMarker))).toBe(false);
 		expect(continuation.messageTexts.some(text => text.includes(planUrl))).toBe(true);
@@ -297,7 +298,7 @@ describe("AgentSession approved-plan reference re-injection after compaction (is
 
 		await session.prompt("do some ordinary work");
 		emitHighUsageTurn(session);
-		const continuation = await waitForCall(call => call.messageTexts.some(text => text.includes(CONTINUE_MARKER)));
+		const continuation = await waitForCall(call => call.callIndex > 0);
 
 		expect(continuation.messageTexts.some(text => text.includes("## Existing Plan"))).toBe(false);
 	});
diff --git a/packages/coding-agent/test/agent-session-prewalk.test.ts b/packages/coding-agent/test/agent-session-prewalk.test.ts
index 7a78ed1d8..6ab6d09f2 100644
--- a/packages/coding-agent/test/agent-session-prewalk.test.ts
+++ b/packages/coding-agent/test/agent-session-prewalk.test.ts
@@ -100,31 +100,13 @@ describe("AgentSession prewalk", () => {
 		return { content: [{ type: "toolCall", id, name, arguments: {} }], stopReason: "toolUse" };
 	}
 
-	function contextMessagesHaveMarker(contextMessages: ReadonlyArray<{ role: string }>, marker: string): boolean {
-		return contextMessages.some(message => {
-			if (message.role !== "user" && message.role !== "developer") return false;
-			if (!("content" in message)) return false;
-			const content: unknown = message.content;
-			if (typeof content === "string") return content.includes(marker);
-			if (!Array.isArray(content)) return false;
-			return content.some(block => {
-				if (typeof block !== "object" || block === null) return false;
-				if (!("type" in block) || block.type !== "text") return false;
-				return "text" in block && typeof block.text === "string" && block.text.includes(marker);
-			});
-		});
-	}
-
 	it("prewalks at the first edit/write after the todo gate opens; bash and todo don't trigger", async () => {
 		const primary = modelOrThrow("claude-sonnet-4-5");
 		const target = modelOrThrow("claude-sonnet-4-6");
 		const modelRegistry = new ModelRegistry(authStorage, path.join(tempDir.path(), "models.yml"));
-		const planMarker = "complete plan in your NEXT reply";
-		const checklistMarker = "grep for every other call site";
 
-		// Turn 1: read-only (nudge injected after). Turn 2: bash — excluded.
-		// Turn 3: todo — opens the gate, must NOT itself switch. Turn 4: write —
-		// first post-todo edit/write, switch.
+		// Turn 1: read-only. Turn 2: bash is excluded. Turn 3: todo opens the gate.
+		// Turn 4: write is the first post-todo edit/write, so it switches.
 		const mock = createMockModel({
 			responses: [
 				toolCall("t1", "record"),
@@ -134,7 +116,7 @@ describe("AgentSession prewalk", () => {
 				{ content: ["done"] },
 			],
 		});
-		const calls: Array<{ model: string; hasNudge: boolean; hasChecklist: boolean }> = [];
+		const calls: string[] = [];
 		const agent = new Agent({
 			getApiKey: () => "test-key",
 			initialState: {
@@ -145,13 +127,9 @@ describe("AgentSession prewalk", () => {
 				thinkingLevel: Effort.Medium,
 			},
 			convertToLlm,
-			streamFn: (model, context, options) => {
-				calls.push({
-					model: `${model.provider}/${model.id}`,
-					hasNudge: contextMessagesHaveMarker(context.messages, planMarker),
-					hasChecklist: contextMessagesHaveMarker(context.messages, checklistMarker),
-				});
-				return mock.stream(model, context, options);
+			streamFn: (model, _context, options) => {
+				calls.push(`${model.provider}/${model.id}`);
+				return mock.stream(model, _context, options);
 			},
 		});
 		session = new AgentSession({
@@ -165,17 +143,13 @@ describe("AgentSession prewalk", () => {
 
 		await session.prompt("do the task");
 
-		expect(calls.map(call => call.model)).toEqual([
+		expect(calls).toEqual([
 			`${primary.provider}/${primary.id}`,
 			`${primary.provider}/${primary.id}`,
 			`${primary.provider}/${primary.id}`,
 			`${primary.provider}/${primary.id}`,
 			`${target.provider}/${target.id}`,
 		]);
-		// Nudge absent on turn 1 (not yet injected), present turns 2-4, scrubbed after the switch.
-		expect(calls.map(call => call.hasNudge)).toEqual([false, true, true, true, false]);
-		// Checklist present only once the target model is running.
-		expect(calls.map(call => call.hasChecklist)).toEqual([false, false, false, false, true]);
 		expect(session.model?.id).toBe(target.id);
 	});
 
@@ -183,11 +157,9 @@ describe("AgentSession prewalk", () => {
 		const primary = modelOrThrow("claude-sonnet-4-5");
 		const target = modelOrThrow("claude-sonnet-4-6");
 		const modelRegistry = new ModelRegistry(authStorage, path.join(tempDir.path(), "models.yml"));
-		const planMarker = "complete plan in your NEXT reply";
 
-		// Turn 1: exploration (nudge after). Turn 2: write with the gate still
-		// closed — no switch; the fast model must not inherit a todo-less run.
-		// Turn 3: todo — gate opens. Turn 4: write — switch.
+		// Turn 1: exploration. Turn 2: write while the gate is closed.
+		// Turn 3: todo opens the gate. Turn 4: write switches.
 		const mock = createMockModel({
 			responses: [
 				toolCall("t1", "record"),
@@ -197,7 +169,7 @@ describe("AgentSession prewalk", () => {
 				{ content: ["done"] },
 			],
 		});
-		const calls: Array<{ model: string; hasNudge: boolean }> = [];
+		const calls: string[] = [];
 		const agent = new Agent({
 			getApiKey: () => "test-key",
 			initialState: {
@@ -208,12 +180,9 @@ describe("AgentSession prewalk", () => {
 				thinkingLevel: Effort.Medium,
 			},
 			convertToLlm,
-			streamFn: (model, context, options) => {
-				calls.push({
-					model: `${model.provider}/${model.id}`,
-					hasNudge: contextMessagesHaveMarker(context.messages, planMarker),
-				});
-				return mock.stream(model, context, options);
+			streamFn: (model, _context, options) => {
+				calls.push(`${model.provider}/${model.id}`);
+				return mock.stream(model, _context, options);
 			},
 		});
 		session = new AgentSession({
@@ -227,15 +196,13 @@ describe("AgentSession prewalk", () => {
 
 		await session.prompt("do the task");
 
-		expect(calls.map(call => call.model)).toEqual([
+		expect(calls).toEqual([
 			`${primary.provider}/${primary.id}`,
 			`${primary.provider}/${primary.id}`,
 			`${primary.provider}/${primary.id}`,
 			`${primary.provider}/${primary.id}`,
 			`${target.provider}/${target.id}`,
 		]);
-		// The turn-2 write landed while the gate was closed — still primary on turn 3.
-		expect(calls.map(call => call.hasNudge)).toEqual([false, true, true, true, false]);
 		expect(session.model?.id).toBe(target.id);
 	});
 
@@ -707,12 +674,10 @@ describe("AgentSession prewalk", () => {
 		).toHaveLength(1);
 	});
 
-	it("armPrewalk rejects a same-model same-effort no-op before injecting the plan nudge", async () => {
+	it("armPrewalk rejects a same-model same-effort no-op", async () => {
 		const model = modelOrThrow("claude-sonnet-4-5");
 		const modelRegistry = new ModelRegistry(authStorage, path.join(tempDir.path(), "models.yml"));
-		const planMarker = "complete plan in your NEXT reply";
 		const mock = createMockModel({ responses: [{ content: ["status only"] }] });
-		const calls: Array<{ hasNudge: boolean }> = [];
 		const agent = new Agent({
 			getApiKey: () => "test-key",
 			initialState: {
@@ -723,10 +688,7 @@ describe("AgentSession prewalk", () => {
 				thinkingLevel: Effort.Medium,
 			},
 			convertToLlm,
-			streamFn: (streamModel, context, options) => {
-				calls.push({ hasNudge: contextMessagesHaveMarker(context.messages, planMarker) });
-				return mock.stream(streamModel, context, options);
-			},
+			streamFn: (streamModel, _context, options) => mock.stream(streamModel, _context, options),
 		});
 		session = new AgentSession({
 			agent,
@@ -744,7 +706,6 @@ describe("AgentSession prewalk", () => {
 		expect(session.armPrewalk(model, Effort.Medium)).toBe(false);
 		await session.prompt("report current status");
 
-		expect(calls).toEqual([{ hasNudge: false }]);
 		expect(notices.some(message => message.includes("nothing to switch"))).toBe(true);
 	});
 
@@ -866,13 +827,11 @@ describe("AgentSession prewalk", () => {
 		// this must still switch.
 		const model = modelOrThrow("claude-sonnet-4-5");
 		const modelRegistry = new ModelRegistry(authStorage, path.join(tempDir.path(), "models.yml"));
-		const checklistMarker = "grep for every other call site";
 
 		// todo excluded from the active slate → the gate opens; record then write.
 		const mock = createMockModel({
 			responses: [toolCall("t1", "record"), toolCall("t2", "write"), { content: ["done"] }],
 		});
-		const calls: Array<{ model: string; hasChecklist: boolean }> = [];
 		const agent = new Agent({
 			getApiKey: () => "test-key",
 			initialState: {
@@ -883,13 +842,7 @@ describe("AgentSession prewalk", () => {
 				thinkingLevel: Effort.Medium,
 			},
 			convertToLlm,
-			streamFn: (streamModel, context, options) => {
-				calls.push({
-					model: `${streamModel.provider}/${streamModel.id}`,
-					hasChecklist: contextMessagesHaveMarker(context.messages, checklistMarker),
-				});
-				return mock.stream(streamModel, context, options);
-			},
+			streamFn: (streamModel, _context, options) => mock.stream(streamModel, _context, options),
 		});
 		session = new AgentSession({
 			agent,
@@ -908,23 +861,16 @@ describe("AgentSession prewalk", () => {
 		// The model id never changes, but the effort drops after the first write.
 		expect(session.model?.id).toBe(model.id);
 		expect(session.thinkingLevel).toBe(Effort.Low);
-		// The switch ran: the post-switch checklist is present on the final turn.
-		expect(calls.at(-1)?.hasChecklist).toBe(true);
 	});
 
-	it("emits a notice and skips the checklist when the prewalk target is a genuine no-op", async () => {
-		// Same model AND same effective thinking level: nothing to switch. The
-		// early return must be visible (a notice), not silent, and must not fire
-		// the post-switch checklist.
+	it("emits a notice when the prewalk target is a genuine no-op", async () => {
+		// Same model and same effective thinking level: no state change.
 		const model = modelOrThrow("claude-sonnet-4-5");
 		const modelRegistry = new ModelRegistry(authStorage, path.join(tempDir.path(), "models.yml"));
-		const planMarker = "complete plan in your NEXT reply";
-		const checklistMarker = "grep for every other call site";
 
 		const mock = createMockModel({
 			responses: [toolCall("t1", "record"), toolCall("t2", "write"), { content: ["done"] }],
 		});
-		const calls: Array<{ hasNudge: boolean; hasChecklist: boolean }> = [];
 		const agent = new Agent({
 			getApiKey: () => "test-key",
 			initialState: {
@@ -935,13 +881,7 @@ describe("AgentSession prewalk", () => {
 				thinkingLevel: Effort.Medium,
 			},
 			convertToLlm,
-			streamFn: (streamModel, context, options) => {
-				calls.push({
-					hasNudge: contextMessagesHaveMarker(context.messages, planMarker),
-					hasChecklist: contextMessagesHaveMarker(context.messages, checklistMarker),
-				});
-				return mock.stream(streamModel, context, options);
-			},
+			streamFn: (streamModel, _context, options) => mock.stream(streamModel, _context, options),
 		});
 		session = new AgentSession({
 			agent,
@@ -963,25 +903,17 @@ describe("AgentSession prewalk", () => {
 		expect(session.thinkingLevel).toBe(Effort.Medium);
 		// The no-op is announced, not silent.
 		expect(notices.some(message => message.includes("nothing to switch"))).toBe(true);
-		// The checklist steer only fires on a real switch.
-		// A genuine no-op must be rejected before the disruptive plan nudge is injected.
-		expect(calls.every(call => !call.hasNudge)).toBe(true);
-		expect(calls.every(call => !call.hasChecklist)).toBe(true);
 	});
 
 	it("treats a target effort the model clamps back to the active effort as a no-op", async () => {
-		// Review edge case: a model capped at `high` running at `high` with a
-		// prewalk target of `xhigh`. The raw selectors differ, but the target
-		// clamps to `high`, so switching would reset the model and inject the
-		// nudges for no effective change — it must be recognized as a no-op.
+		// A model capped at high resolves an xhigh target back to high.
+		// The equal effective settings must be recognized as a no-op.
 		const model = modelOrThrow("claude-sonnet-4-6"); // supported efforts cap at high
 		const modelRegistry = new ModelRegistry(authStorage, path.join(tempDir.path(), "models.yml"));
-		const checklistMarker = "grep for every other call site";
 
 		const mock = createMockModel({
 			responses: [toolCall("t1", "record"), toolCall("t2", "write"), { content: ["done"] }],
 		});
-		const calls: Array<{ hasChecklist: boolean }> = [];
 		const agent = new Agent({
 			getApiKey: () => "test-key",
 			initialState: {
@@ -992,10 +924,7 @@ describe("AgentSession prewalk", () => {
 				thinkingLevel: Effort.High,
 			},
 			convertToLlm,
-			streamFn: (streamModel, context, options) => {
-				calls.push({ hasChecklist: contextMessagesHaveMarker(context.messages, checklistMarker) });
-				return mock.stream(streamModel, context, options);
-			},
+			streamFn: (streamModel, _context, options) => mock.stream(streamModel, _context, options),
 		});
 		session = new AgentSession({
 			agent,
@@ -1015,7 +944,6 @@ describe("AgentSession prewalk", () => {
 
 		expect(session.thinkingLevel).toBe(Effort.High);
 		expect(notices.some(message => message.includes("nothing to switch"))).toBe(true);
-		expect(calls.every(call => !call.hasChecklist)).toBe(true);
 	});
 
 	it("switches when a same-model target clears auto mode even though efforts both resolve to undefined", async () => {
@@ -1025,12 +953,10 @@ describe("AgentSession prewalk", () => {
 		// collapse to a no-op.
 		const model = modelOrThrow("claude-sonnet-4-5");
 		const modelRegistry = new ModelRegistry(authStorage, path.join(tempDir.path(), "models.yml"));
-		const checklistMarker = "grep for every other call site";
 
 		const mock = createMockModel({
 			responses: [toolCall("t1", "record"), toolCall("t2", "write"), { content: ["done"] }],
 		});
-		const calls: Array<{ hasChecklist: boolean }> = [];
 		const agent = new Agent({
 			getApiKey: () => "test-key",
 			initialState: {
@@ -1041,10 +967,7 @@ describe("AgentSession prewalk", () => {
 				thinkingLevel: Effort.Medium,
 			},
 			convertToLlm,
-			streamFn: (streamModel, context, options) => {
-				calls.push({ hasChecklist: contextMessagesHaveMarker(context.messages, checklistMarker) });
-				return mock.stream(streamModel, context, options);
-			},
+			streamFn: (streamModel, _context, options) => mock.stream(streamModel, _context, options),
 		});
 		session = new AgentSession({
 			agent,
@@ -1064,9 +987,8 @@ describe("AgentSession prewalk", () => {
 
 		await session.prompt("do the task");
 
-		// The hand-off ran: auto is cleared and the post-switch checklist fired.
+		// The hand-off clears automatic thinking.
 		expect(session.isAutoThinking).toBe(false);
 		expect(notices.some(message => message.includes("nothing to switch"))).toBe(false);
-		expect(calls.at(-1)?.hasChecklist).toBe(true);
 	});
 });
diff --git a/packages/coding-agent/test/interactive-mode-vibe-toggle.test.ts b/packages/coding-agent/test/interactive-mode-vibe-toggle.test.ts
index d9f6448c8..edc3ce00d 100644
--- a/packages/coding-agent/test/interactive-mode-vibe-toggle.test.ts
+++ b/packages/coding-agent/test/interactive-mode-vibe-toggle.test.ts
@@ -17,7 +17,6 @@ import { InteractiveMode } from "@oh-my-pi/pi-coding-agent/modes/interactive-mod
 import { initTheme } from "@oh-my-pi/pi-coding-agent/modes/theme/theme";
 import { AgentSession } from "@oh-my-pi/pi-coding-agent/session/agent-session";
 import { AuthStorage } from "@oh-my-pi/pi-coding-agent/session/auth-storage";
-import { normalizeCustomMessagePayload } from "@oh-my-pi/pi-coding-agent/session/messages";
 import { SessionManager } from "@oh-my-pi/pi-coding-agent/session/session-manager";
 import { FileSessionStorage, type WriteTextAtomicOptions } from "@oh-my-pi/pi-coding-agent/session/session-storage";
 import { VIBE_TOOL_NAMES } from "@oh-my-pi/pi-coding-agent/tools/vibe";
@@ -150,13 +149,6 @@ describe("InteractiveMode vibe mode toggle", () => {
 		expect(inMode.toSorted()).toEqual(["read", "todo", ...VIBE_TOOL_NAMES].toSorted());
 		expect(session.getAllToolNames().toSorted()).toEqual(["read", "todo", ...VIBE_TOOL_NAMES].toSorted());
 
-		const sendCustomMessage = vi.spyOn(session, "sendCustomMessage");
-		await session.sendVibeModeContext({ deliverAs: "steer" });
-		const message = normalizeCustomMessagePayload(sendCustomMessage.mock.calls[0]?.[0]);
-		const content = typeof message.content === "string" ? message.content : "";
-		expect(message.customType).toBe("vibe-mode-context");
-		expect(content).toContain("`todo`");
-
 		// Toggle off: the empty previous toolset must come back — only the
 		// ephemeral vibe tools must leave the registry.
 		await mode.handleVibeModeCommand();
@@ -198,13 +190,6 @@ describe("InteractiveMode vibe mode toggle", () => {
 			await foreignTodoMode.handleVibeModeCommand();
 			expect(foreignTodoSession.getActiveToolNames().toSorted()).toEqual(["read", ...VIBE_TOOL_NAMES].toSorted());
 
-			const sendCustomMessage = vi.spyOn(foreignTodoSession, "sendCustomMessage");
-			await foreignTodoSession.sendVibeModeContext({ deliverAs: "steer" });
-			const message = normalizeCustomMessagePayload(sendCustomMessage.mock.calls[0]?.[0]);
-			const content = typeof message.content === "string" ? message.content : "";
-			expect(content).not.toContain("`todo`");
-			expect(content).not.toContain("parent session's list");
-
 			await foreignTodoMode.handleVibeModeCommand();
 			expect(foreignTodoSession.getActiveToolNames()).toEqual([]);
 			expect(foreignTodoSession.getAllToolNames().toSorted()).toEqual(["read", "todo"]);
@@ -234,12 +219,6 @@ describe("InteractiveMode vibe mode toggle", () => {
 
 		expect(mode.vibeModeEnabled).toBe(true);
 		expect(session.getActiveToolNames()).toEqual(expect.arrayContaining(["read", "todo", ...VIBE_TOOL_NAMES]));
-		const sendCustomMessage = vi.spyOn(session, "sendCustomMessage");
-		await session.sendVibeModeContext({ deliverAs: "steer" });
-		const message = normalizeCustomMessagePayload(sendCustomMessage.mock.calls[0]?.[0]);
-		const content = typeof message.content === "string" ? message.content : "";
-		expect(content).toContain("`todo`");
-		expect(content).toContain("parent session's list");
 		expect(suspend).toHaveBeenCalledTimes(1);
 		expect(terminate).not.toHaveBeenCalled();
 		expect(vibeModeEntryCount(session.sessionManager)).toBe(1);
diff --git a/packages/coding-agent/test/issue-8223-repro.test.ts b/packages/coding-agent/test/issue-8223-repro.test.ts
index dca3eb145..17d32537d 100644
--- a/packages/coding-agent/test/issue-8223-repro.test.ts
+++ b/packages/coding-agent/test/issue-8223-repro.test.ts
@@ -60,16 +60,11 @@ test("keeps Gemini 3.6 advisor context and accepts a silent review", async () =>
 		expect(bodies).toHaveLength(1);
 		expect(bodies[0]).toMatchObject({
 			systemInstruction: {
-				parts: [{ text: expect.stringContaining("You bring a different angle") }],
+				parts: [{ text: expect.any(String) }],
 			},
 			tools: [
 				{
-					functionDeclarations: [
-						{
-							name: "advise",
-							description: expect.stringContaining("Send one concrete"),
-						},
-					],
+					functionDeclarations: [{ name: "advise" }],
 				},
 			],
 		});
diff --git a/packages/coding-agent/test/modes/workflow.test.ts b/packages/coding-agent/test/modes/workflow.test.ts
index 9bab3f671..398587cb2 100644
--- a/packages/coding-agent/test/modes/workflow.test.ts
+++ b/packages/coding-agent/test/modes/workflow.test.ts
@@ -1,11 +1,6 @@
 import { beforeAll, describe, expect, it } from "bun:test";
 import { initTheme } from "@oh-my-pi/pi-coding-agent/modes/theme/theme";
-import {
-	containsWorkflow,
-	highlightWorkflow,
-	renderWorkflowNotice,
-	WORKFLOW_NOTICE,
-} from "@oh-my-pi/pi-coding-agent/modes/workflow";
+import { containsWorkflow, highlightWorkflow } from "@oh-my-pi/pi-coding-agent/modes/workflow";
 
 beforeAll(() => {
 	// highlightWorkflow reads the global theme's color mode.
@@ -63,17 +58,3 @@ describe("workflow keyword highlighting", () => {
 		expect(highlightWorkflow(filePath)).toBe(filePath);
 	});
 });
-
-describe("workflow notice", () => {
-	it("renders the Workflowz trigger with eval orchestration helper guidance", () => {
-		expect(WORKFLOW_NOTICE).toContain("**workflowz** keyword");
-		expect(WORKFLOW_NOTICE).toContain("`parallel(thunks)`");
-		expect(WORKFLOW_NOTICE).toContain("await budget.remaining()");
-	});
-
-	it("renders the same eval notice when task.batch is disabled", () => {
-		const notice = renderWorkflowNotice({ taskBatch: false });
-		expect(notice).toContain("**workflowz** keyword");
-		expect(notice).toContain("`parallel(thunks)`");
-	});
-});
diff --git a/packages/coding-agent/test/system-prompt-personality.test.ts b/packages/coding-agent/test/system-prompt-personality.test.ts
deleted file mode 100644
index 1706f07a1..000000000
--- a/packages/coding-agent/test/system-prompt-personality.test.ts
+++ /dev/null
@@ -1,63 +0,0 @@
-import { afterEach, beforeEach, describe, expect, it } from "bun:test";
-import * as fs from "node:fs";
-import * as os from "node:os";
-import * as path from "node:path";
-import type { Personality } from "@oh-my-pi/pi-coding-agent/config/settings-schema";
-import { buildSystemPrompt } from "@oh-my-pi/pi-coding-agent/system-prompt";
-import { cleanupTempHome } from "./helpers/temp-home-cleanup";
-
-const EMPTY_TREE = {
-	rootPath: "",
-	rendered: "",
-	truncated: false,
-	totalLines: 0,
-	agentsMdFiles: [],
-};
-
-describe("system prompt personality block", () => {
-	let tempDir = "";
-	let tempHomeDir = "";
-	let originalHome: string | undefined;
-
-	beforeEach(() => {
-		tempDir = fs.mkdtempSync(path.join(os.tmpdir(), "pi-prompt-personality-"));
-		tempHomeDir = fs.mkdtempSync(path.join(os.tmpdir(), "pi-prompt-personality-home-"));
-		originalHome = process.env.HOME;
-		process.env.HOME = tempHomeDir;
-	});
-
-	afterEach(cleanupTempHome(() => ({ tempDir, tempHomeDir, originalHome })));
-
-	async function render(personality?: Personality): Promise<string> {
-		const { systemPrompt } = await buildSystemPrompt({
-			cwd: tempDir,
-			contextFiles: [],
-			skills: [],
-			rules: [],
-			toolNames: [],
-			workspaceTree: { ...EMPTY_TREE, rootPath: tempDir },
-			personality,
-		});
-		return systemPrompt.join("\n\n");
-	}
-
-	it("injects the default personality when the option is unset", async () => {
-		const rendered = await render();
-		expect(rendered).toContain("<personality>");
-		expect(rendered).toContain("</personality>");
-		expect(rendered).toContain("terse, evidence-first engineer");
-	});
-
-	it("replaces the default spec when a non-default personality is selected", async () => {
-		const rendered = await render("friendly");
-		expect(rendered).toContain("<personality>");
-		expect(rendered).toContain("warm, supportive collaborator");
-		expect(rendered).not.toContain("terse, evidence-first engineer");
-	});
-
-	it('omits the personality block entirely for "none"', async () => {
-		const rendered = await render("none");
-		expect(rendered).not.toContain("<personality>");
-		expect(rendered).not.toContain("</personality>");
-	});
-});
diff --git a/packages/coding-agent/test/task/executor-soft-budget.test.ts b/packages/coding-agent/test/task/executor-soft-budget.test.ts
index 5571bd89c..d0c778899 100644
--- a/packages/coding-agent/test/task/executor-soft-budget.test.ts
+++ b/packages/coding-agent/test/task/executor-soft-budget.test.ts
@@ -259,9 +259,8 @@ describe("runSubprocess soft request budget", () => {
 		// wrap-up reminder; the second abort (after the terminal yield) is the
 		// normal post-yield terminate.
 		expect(abortCallsAtReminder).toBe(1);
-		// The wrap-up reminder is the budget-stop variant with a forced tool choice.
+		// The budget stop forces a synthetic terminal yield.
 		expect(handle.prompts).toHaveLength(2);
-		expect(handle.prompts[1]?.text).toMatch(/request budget/);
 		expect(handle.prompts[1]?.options?.synthetic).toBe(true);
 		expect(handle.prompts[1]?.options?.toolChoice).toEqual({ type: "tool", name: "yield" });
 		// The forced yield finalizes as a normal completion, not an abort.
diff --git a/packages/coding-agent/test/task/executor-subagent-reminders.test.ts b/packages/coding-agent/test/task/executor-subagent-reminders.test.ts
index d3babdff9..bdfea201f 100644
--- a/packages/coding-agent/test/task/executor-subagent-reminders.test.ts
+++ b/packages/coding-agent/test/task/executor-subagent-reminders.test.ts
@@ -234,7 +234,7 @@ describe("runSubprocess yield reminders", () => {
 		expect(systemPrompt).toHaveLength(4);
 		expect(systemPrompt?.[0]).toBe("system");
 		expect(systemPrompt?.[1]).toBe("project");
-		expect(systemPrompt?.[2]).toMatch(/ROLE\n=+\n\ntest/);
+		expect(systemPrompt?.[2]).toContain(baseAgent.systemPrompt);
 		// The parent-conversation CONTEXT section is gone: subagents get their
 		// background inside the assignment (or a local:// file), never a dump.
 		expect(systemPrompt?.[2]).not.toMatch(/CONTEXT\n=+/);
diff --git a/packages/coding-agent/test/task/spawn-policy.test.ts b/packages/coding-agent/test/task/spawn-policy.test.ts
index 17c2cd01b..f963733bc 100644
--- a/packages/coding-agent/test/task/spawn-policy.test.ts
+++ b/packages/coding-agent/test/task/spawn-policy.test.ts
@@ -1,6 +1,5 @@
 import { afterEach, describe, expect, it, vi } from "bun:test";
 import { Settings } from "../../src/config/settings";
-import initAgentPrompt from "../../src/prompts/agents/init.md" with { type: "text" };
 import * as taskDiscovery from "../../src/task/discovery";
 import { TaskTool } from "../../src/task/index";
 import { isScoutSpawnable } from "../../src/task/spawn-policy";
@@ -131,10 +130,3 @@ describe("task tool description scout gating", () => {
 		expect(description).toContain("### reviewer");
 	});
 });
-
-describe("bundled agent prompt scout gating", () => {
-	it("does not hard-code scout in the init agent prompt", () => {
-		expect(initAgentPrompt.toLowerCase()).not.toContain("scout");
-		expect(initAgentPrompt).toContain("multiple research agents");
-	});
-});