@oh-my-pi/pi-coding-agent 14.5.8 → 14.5.10

package/src/prompts/tools/todo-write.md

@@ -1,69 +1,33 @@
-Manages a phased task list through an `ops` array of flat operations.
-The next pending task is auto-promoted to `in_progress` after completing the current one.
+Manages a phased task list. Pass `ops`: a flat array of operations.
+The next pending task is auto-promoted to `in_progress` after each completion.
 
-<protocol>
-## Shape
+## Operations
 
-Pass an object with an `ops` array:
-
-```ts
-{
-  ops: [
-    { op: "replace", phases: [...] },
-    { op: "start", task: "task-3" },
-    { op: "done", phase: "Implementation" },
-    { op: "rm" },
-    { op: "drop", task: "task-9" },
-    { op: "append", phase: "Implementation", items: [{ id: "task-10", label: "Run tests" }] },
-  ],
-}
-```
-
-## Operation fields
-
-|Field|Type|When to use|
+|`op`|Required fields|Effect|
 |---|---|---|
-|`op`|string|Required. One of `replace`, `start`, `done`, `rm`, `drop`, `append`, `note`|
-|`task`|string|Task id for `start`, or a task target for `done` / `rm` / `drop`|
-|`phase`|string|Phase target for `done` / `rm` / `drop`, or append destination for `append`|
-|`items`|{id, label}[]|Required for `append`. If the phase does not exist, it is created at the end|
-|`phases`|Phase[]|Only for `replace`. Keeps initial phased setup available for harness bootstrap and full restructures|
-|`text`|string|Required for `note`. The note text appended to `task.notes` (which is a list, joined with newlines on render)|
-
-## Semantics
-- `start`: requires `task`; sets that task to `in_progress`
-- `done`: marks one task, one phase, or all tasks completed
-- `rm`: removes one task, one phase's tasks, or all tasks
-- `drop`: marks one task, one phase, or all tasks abandoned
-- `append`: appends `items` to `phase`; creates the phase if missing
-- `replace`: replaces the full todo list
-- `note`: append `text` as a new note attached to `task`. Notes are append-only context the user added; they only render to you when the task is `in_progress`. Other tasks display only a `+N` marker. Use this when you want to leave a follow-up reminder for yourself when you reach a later task.
-
-If `done`, `rm`, or `drop` omits both `task` and `phase`, it applies to all tasks.
-
-## Task Anatomy
-- `label`: Short label (5-10 words). What is being done, not how.
-- `replace` task `content` should stay short and specific.
-
-## Phase Anatomy
-- `name`: Short, human-readable noun phrase (1-3 words). Capitalize naturally.
-- Always prefix with a roman-numeral ordinal (`I.`, `II.`, `III.`, `IV.`, …) to convey ordering — e.g. `I. Foundation`, `II. Auth`, `III. Routing`. Single-phase plans use `I.` too.
-- You **MUST NOT** use snake_case, `Phase1_*`, arabic numerals (`1.`), or letter prefixes (`A.`) — they render as ugly identifiers.
+|`replace`|`phases`|Replace the full list (initial setup, full restructure)|
+|`start`|`task`|Set task to `in_progress`|
+|`done`|`task` or `phase` (or neither = all)|Mark completed|
+|`drop`|`task` or `phase` (or neither = all)|Mark abandoned|
+|`rm`|`task` or `phase` (or neither = all)|Remove|
+|`append`|`phase`, `items: {id, label}[]`|Append tasks; creates phase if missing|
+|`note`|`task`, `text`|Append a note to `task.notes`. Only use to leave reminders for future-you.|
+
+## Anatomy
+- **Task `label`**: 5–10 words, what is being done, not how.
+- **Phase `name`**: short noun phrase prefixed with a roman numeral — `I. Foundation`, `II. Auth`, `III. Verification`. Single-phase plans still use `I.`. Never use snake_case, arabic numerals, or letter prefixes.
 
 ## Rules
 - Mark tasks done immediately after finishing — never defer.
-- Complete phases in order — do not skip ahead while earlier ones are pending.
-- On blockers, append a new task to the active phase.
+- Complete phases in order.
+- On blockers, `append` a new task to the active phase.
 - Keep ids stable once introduced.
-</protocol>
 
-<conditions>
-Create a todo list when:
-1. Task requires 3+ distinct steps
-2. User explicitly requests one
-3. User provides a set of tasks to complete
-4. New instructions arrive mid-task — capture before proceeding
-</conditions>
+## When to create a list
+- Task requires 3+ distinct steps
+- User explicitly requests one
+- User provides a set of tasks to complete
+- New instructions arrive mid-task — capture before proceeding
 
 <examples>
 # Initial setup (multi-phase)
@@ -81,9 +45,3 @@ Create a todo list when:
 # Append tasks to a phase
 `{"ops":[{"op":"append","phase":"II. Auth","items":[{"id":"task-8","label":"Handle retries"},{"id":"task-9","label":"Run tests"}]}]}`
 </examples>
-
-<avoid>
-- Single-step tasks — act directly
-- Conversational or informational requests
-- Tasks completable in under 3 trivial steps
-</avoid>

package/src/sdk.ts

@@ -6,7 +6,7 @@ import {
 	INTENT_FIELD,
 	type ThinkingLevel,
 } from "@oh-my-pi/pi-agent-core";
-import type { Message, Model } from "@oh-my-pi/pi-ai";
+import type { Message, Model, SimpleStreamOptions } from "@oh-my-pi/pi-ai";
 import {
 	getOpenAICodexTransportDetails,
 	prewarmOpenAICodexResponses,
@@ -793,7 +793,11 @@ export async function createAgentSession(options: CreateAgentSessionOptions = {}
 		thinkingLevel = defaultRoleSpec.thinkingLevel;
 	}
 
-	// Fall back to settings default
+	// Prefer the selected model's configured defaultLevel, otherwise fall back
+	// to the global settings default.
+	if (thinkingLevel === undefined && model?.thinking?.defaultLevel !== undefined) {
+		thinkingLevel = model.thinking.defaultLevel;
+	}
 	if (thinkingLevel === undefined) {
 		thinkingLevel = settings.get("defaultThinkingLevel");
 	}
@@ -1498,6 +1502,11 @@ export async function createAgentSession(options: CreateAgentSessionOptions = {}
 					return await extensionRunner.emitBeforeProviderRequest(payload);
 				}
 			: undefined;
+		const onResponse: SimpleStreamOptions["onResponse"] | undefined = extensionRunner
+			? async (response, model) => {
+					await extensionRunner.emitAfterProviderResponse(response, model);
+				}
+			: undefined;
 
 		const setToolUIContext = (uiContext: ExtensionUIContext, hasUI: boolean) => {
 			toolContextStore.setUIContext(uiContext, hasUI);
@@ -1527,6 +1536,7 @@ export async function createAgentSession(options: CreateAgentSessionOptions = {}
 			},
 			convertToLlm: convertToLlmFinal,
 			onPayload,
+			onResponse,
 			sessionId: providerSessionId,
 			transformContext,
 			steeringMode: settings.get("steeringMode") ?? "one-at-a-time",
@@ -1599,6 +1609,7 @@ export async function createAgentSession(options: CreateAgentSessionOptions = {}
 			toolRegistry,
 			transformContext,
 			onPayload,
+			onResponse,
 			convertToLlm: convertToLlmFinal,
 			rebuildSystemPrompt,
 			mcpDiscoveryEnabled,

package/src/session/agent-session.ts

@@ -46,6 +46,7 @@ import {
 	calculateRateLimitBackoffMs,
 	getSupportedEfforts,
 	isContextOverflow,
+	isUnexpectedSocketCloseMessage,
 	isUsageLimitError,
 	modelsAreEqual,
 	parseRateLimitReason,
@@ -104,7 +105,7 @@ import { ExtensionToolWrapper } from "../extensibility/extensions/wrapper";
 import type { HookCommandContext } from "../extensibility/hooks/types";
 import type { Skill, SkillWarning } from "../extensibility/skills";
 import { expandSlashCommand, type FileSlashCommand } from "../extensibility/slash-commands";
-import { resolveLocalUrlToPath } from "../internal-urls";
+import { type LocalProtocolOptions, resolveLocalUrlToPath } from "../internal-urls";
 import {
 	disposeKernelSessionsByOwner,
 	executePython as executePythonCommand,
@@ -120,6 +121,7 @@ import {
 } from "../mcp/discoverable-tool-metadata";
 import { getCurrentThemeName, theme } from "../modes/theme/theme";
 import type { PlanModeState } from "../plan-mode/state";
+import autoContinuePrompt from "../prompts/system/auto-continue.md" with { type: "text" };
 import autoHandoffThresholdFocusPrompt from "../prompts/system/auto-handoff-threshold-focus.md" with { type: "text" };
 import eagerTodoPrompt from "../prompts/system/eager-todo.md" with { type: "text" };
 import handoffDocumentPrompt from "../prompts/system/handoff-document.md" with { type: "text" };
@@ -244,6 +246,8 @@ export interface AgentSessionConfig {
 	transformContext?: (messages: AgentMessage[], signal?: AbortSignal) => AgentMessage[] | Promise<AgentMessage[]>;
 	/** Provider payload hook used by the active session request path */
 	onPayload?: SimpleStreamOptions["onPayload"];
+	/** Provider response hook used by the active session request path */
+	onResponse?: SimpleStreamOptions["onResponse"];
 	/** Current session message-to-LLM conversion pipeline */
 	convertToLlm?: (messages: AgentMessage[]) => Message[] | Promise<Message[]>;
 	/** System prompt builder that can consider tool availability */
@@ -469,7 +473,7 @@ export class AgentSession {
 	#toolChoiceQueue = new ToolChoiceQueue();
 
 	// Bash execution state
-	#bashAbortController: AbortController | undefined = undefined;
+	#bashAbortControllers = new Set<AbortController>();
 	#pendingBashMessages: BashExecutionMessage[] = [];
 
 	// Python execution state
@@ -507,6 +511,7 @@ export class AgentSession {
 	#toolRegistry: Map<string, AgentTool>;
 	#transformContext: (messages: AgentMessage[], signal?: AbortSignal) => AgentMessage[] | Promise<AgentMessage[]>;
 	#onPayload: SimpleStreamOptions["onPayload"] | undefined;
+	#onResponse: SimpleStreamOptions["onResponse"] | undefined;
 	#convertToLlm: (messages: AgentMessage[]) => Message[] | Promise<Message[]>;
 	#rebuildSystemPrompt: ((toolNames: string[], tools: Map<string, AgentTool>) => Promise<string>) | undefined;
 	#baseSystemPrompt: string;
@@ -526,8 +531,7 @@ export class AgentSession {
 	#ttsrRetryToken = 0;
 	#ttsrResumePromise: Promise<void> | undefined = undefined;
 	#ttsrResumeResolve: (() => void) | undefined = undefined;
-	#postPromptTaskCounter = 0;
-	#postPromptTaskIds = new Set<number>();
+	#postPromptTasks = new Set<Promise<void>>();
 	#postPromptTasksPromise: Promise<void> | undefined = undefined;
 	#postPromptTasksResolve: (() => void) | undefined = undefined;
 	#postPromptTasksAbortController = new AbortController();
@@ -593,6 +597,7 @@ export class AgentSession {
 		this.#toolRegistry = config.toolRegistry ?? new Map();
 		this.#transformContext = config.transformContext ?? (messages => messages);
 		this.#onPayload = config.onPayload;
+		this.#onResponse = config.onResponse;
 		this.#convertToLlm = config.convertToLlm ?? convertToLlm;
 		this.#rebuildSystemPrompt = config.rebuildSystemPrompt;
 		this.#baseSystemPrompt = this.agent.state.systemPrompt;
@@ -1144,14 +1149,13 @@ export class AgentSession {
 	}
 
 	#trackPostPromptTask(task: Promise<void>): void {
-		const taskId = ++this.#postPromptTaskCounter;
-		this.#postPromptTaskIds.add(taskId);
+		this.#postPromptTasks.add(task);
 		this.#ensurePostPromptTasksPromise();
 		void task
 			.catch(() => {})
 			.finally(() => {
-				this.#postPromptTaskIds.delete(taskId);
-				if (this.#postPromptTaskIds.size === 0) {
+				this.#postPromptTasks.delete(task);
+				if (this.#postPromptTasks.size === 0) {
 					this.#resolvePostPromptTasks();
 				}
 			});
@@ -1217,11 +1221,11 @@ export class AgentSession {
 			await this.#promptWithMessage(
 				{
 					role: "developer",
-					content: [{ type: "text", text: "Continue if you have next steps." }],
+					content: [{ type: "text", text: autoContinuePrompt }],
 					attribution: "agent",
 					timestamp: Date.now(),
 				},
-				"Continue if you have next steps.",
+				autoContinuePrompt,
 				{ skipPostPromptRecoveryWait: true },
 			);
 		};
@@ -1235,11 +1239,21 @@ export class AgentSession {
 		);
 	}
 
-	#cancelPostPromptTasks(): void {
+	async #cancelPostPromptTasks(): Promise<void> {
 		this.#postPromptTasksAbortController.abort();
 		this.#postPromptTasksAbortController = new AbortController();
-		this.#postPromptTaskIds.clear();
-		this.#resolvePostPromptTasks();
+		this.#resolveTtsrResume();
+
+		const pendingTasks = Array.from(this.#postPromptTasks);
+		if (pendingTasks.length === 0) {
+			this.#resolvePostPromptTasks();
+			return;
+		}
+
+		await Promise.allSettled(pendingTasks);
+		if (this.#postPromptTasks.size === 0) {
+			this.#resolvePostPromptTasks();
+		}
 	}
 	/**
 	 * Wait for retry, TTSR resume, and any background continuation to settle.
@@ -1523,10 +1537,19 @@ export class AgentSession {
 		const path = typeof args.path === "string" ? args.path : undefined;
 		if (!path) return undefined;
 
+		// `local://` URLs (e.g. local://PLAN.md for plan-mode) resolve to a real
+		// on-disk artifacts path; pre-caching works as long as we ask the
+		// local-protocol handler. Other internal-scheme URLs (agent://, skill://,
+		// rule://, mcp://, artifact://) have no stable filesystem representation;
+		// skip pre-cache entirely for those — the edit tool itself will reject
+		// them through its normal dispatch path.
+		const resolvedPath = this.#resolveSessionFsPath(path);
+		if (resolvedPath === undefined) return undefined;
+
 		return {
 			toolCall,
 			path,
-			resolvedPath: resolveToCwd(path, this.sessionManager.getCwd()),
+			resolvedPath,
 			diff: typeof args.diff === "string" ? args.diff : undefined,
 			op: typeof args.op === "string" ? args.op : undefined,
 			rename: typeof args.rename === "string" ? args.rename : undefined,
@@ -1600,11 +1623,47 @@ export class AgentSession {
 	}
 
 	/** Invalidate cache for a file after an edit completes to prevent stale data */
-	#invalidateFileCacheForPath(path: string): void {
-		const resolvedPath = resolveToCwd(path, this.sessionManager.getCwd());
+	#invalidateFileCacheForPath(filePath: string): void {
+		const resolvedPath = this.#resolveSessionFsPath(filePath);
+		if (resolvedPath === undefined) return;
 		this.#streamingEditFileCache.delete(resolvedPath);
 	}
 
+	/**
+	 * Resolve a path supplied to a tool to a real filesystem path.
+	 *
+	 * - `local://` URLs route through the local-protocol handler so they map
+	 *   onto the session's on-disk artifacts directory; pre-caching, ENOENT
+	 *   handling, and post-edit invalidation all work normally.
+	 * - Other internal-scheme URLs (agent://, skill://, rule://, mcp://,
+	 *   artifact://) have no stable filesystem path; this returns `undefined`
+	 *   so callers skip filesystem-only operations.
+	 * - Cwd-relative and absolute paths resolve via `resolveToCwd`.
+	 */
+	#resolveSessionFsPath(filePath: string): string | undefined {
+		const normalized = normalizeLocalScheme(filePath);
+		if (normalized.startsWith("local:")) {
+			return resolveLocalUrlToPath(normalized, this.#localProtocolOptions());
+		}
+		if (
+			normalized.startsWith("agent://") ||
+			normalized.startsWith("skill://") ||
+			normalized.startsWith("rule://") ||
+			normalized.startsWith("mcp://") ||
+			normalized.startsWith("artifact://")
+		) {
+			return undefined;
+		}
+		return resolveToCwd(normalized, this.sessionManager.getCwd());
+	}
+
+	#localProtocolOptions(): LocalProtocolOptions {
+		return {
+			getArtifactsDir: () => this.sessionManager.getArtifactsDir(),
+			getSessionId: () => this.sessionManager.getSessionId(),
+		};
+	}
+
 	#maybeAbortStreamingEdit(event: AgentEvent): void {
 		if (!this.settings.get("edit.streamingAbort")) return;
 		if (this.#streamingEditAbortTriggered) return;
@@ -1892,7 +1951,7 @@ export class AgentSession {
 		} catch (error) {
 			logger.warn("Failed to emit session_shutdown event", { error: String(error) });
 		}
-		this.#cancelPostPromptTasks();
+		await this.#cancelPostPromptTasks();
 		this.#clearTodoClearTimers();
 		const drained = await this.#asyncJobManager?.dispose({ timeoutMs: 3_000 });
 		const deliveryState = this.#asyncJobManager?.getDeliveryState();
@@ -2318,21 +2377,39 @@ export class AgentSession {
 
 	/** Apply session-level stream hooks to a direct side request. */
 	prepareSimpleStreamOptions(options: SimpleStreamOptions): SimpleStreamOptions {
-		if (!this.#onPayload) return options;
-		if (!options.onPayload) {
-			return { ...options, onPayload: this.#onPayload };
-		}
 		const sessionOnPayload = this.#onPayload;
-		const requestOnPayload = options.onPayload;
-		return {
-			...options,
-			onPayload: async (payload, model) => {
-				const sessionPayload = await sessionOnPayload(payload, model);
-				const sessionResolvedPayload = sessionPayload ?? payload;
-				const requestPayload = await requestOnPayload(sessionResolvedPayload, model);
-				return requestPayload ?? sessionResolvedPayload;
-			},
-		};
+		const sessionOnResponse = this.#onResponse;
+		if (!sessionOnPayload && !sessionOnResponse) return options;
+
+		const preparedOptions: SimpleStreamOptions = { ...options };
+
+		if (sessionOnPayload) {
+			if (!options.onPayload) {
+				preparedOptions.onPayload = sessionOnPayload;
+			} else {
+				const requestOnPayload = options.onPayload;
+				preparedOptions.onPayload = async (payload, model) => {
+					const sessionPayload = await sessionOnPayload(payload, model);
+					const sessionResolvedPayload = sessionPayload ?? payload;
+					const requestPayload = await requestOnPayload(sessionResolvedPayload, model);
+					return requestPayload ?? sessionResolvedPayload;
+				};
+			}
+		}
+
+		if (sessionOnResponse) {
+			if (!options.onResponse) {
+				preparedOptions.onResponse = sessionOnResponse;
+			} else {
+				const requestOnResponse = options.onResponse;
+				preparedOptions.onResponse = async (response, model) => {
+					await sessionOnResponse(response, model);
+					await requestOnResponse(response, model);
+				};
+			}
+		}
+
+		return preparedOptions;
 	}
 
 	/** Current steering mode */
@@ -2466,10 +2543,7 @@ export class AgentSession {
 		if (this.#planReferenceSent) return null;
 
 		const planFilePath = this.#planReferencePath;
-		const resolvedPlanPath = resolveLocalUrlToPath(planFilePath, {
-			getArtifactsDir: () => this.sessionManager.getArtifactsDir(),
-			getSessionId: () => this.sessionManager.getSessionId(),
-		});
+		const resolvedPlanPath = resolveLocalUrlToPath(planFilePath, this.#localProtocolOptions());
 		let planContent: string;
 		try {
 			planContent = await Bun.file(resolvedPlanPath).text();
@@ -2502,15 +2576,9 @@ export class AgentSession {
 		if (!state?.enabled) return null;
 		const sessionPlanUrl = "local://PLAN.md";
 		const resolvedPlanPath = state.planFilePath.startsWith("local:")
-			? resolveLocalUrlToPath(normalizeLocalScheme(state.planFilePath), {
-					getArtifactsDir: () => this.sessionManager.getArtifactsDir(),
-					getSessionId: () => this.sessionManager.getSessionId(),
-				})
+			? resolveLocalUrlToPath(normalizeLocalScheme(state.planFilePath), this.#localProtocolOptions())
 			: resolveToCwd(state.planFilePath, this.sessionManager.getCwd());
-		const resolvedSessionPlan = resolveLocalUrlToPath(sessionPlanUrl, {
-			getArtifactsDir: () => this.sessionManager.getArtifactsDir(),
-			getSessionId: () => this.sessionManager.getSessionId(),
-		});
+		const resolvedSessionPlan = resolveLocalUrlToPath(sessionPlanUrl, this.#localProtocolOptions());
 		const displayPlanPath =
 			state.planFilePath.startsWith("local:") || resolvedPlanPath !== resolvedSessionPlan
 				? state.planFilePath
@@ -3358,9 +3426,13 @@ export class AgentSession {
 		this.abortRetry();
 		this.#promptGeneration++;
 		this.#scheduledHiddenNextTurnGeneration = undefined;
-		this.#resolveTtsrResume();
-		this.#cancelPostPromptTasks();
+		this.abortCompaction();
+		this.abortHandoff();
+		this.abortBash();
+		this.abortPython();
+		const postPromptDrain = this.#cancelPostPromptTasks();
 		this.agent.abort();
+		await postPromptDrain;
 		await this.agent.waitForIdle();
 		// Clear prompt-in-flight state: waitForIdle resolves when the agent loop's finally
 		// block runs, but nested prompt setup/finalizers may still be unwinding. Without this,
@@ -3555,8 +3627,9 @@ export class AgentSession {
 		);
 		this.settings.getStorage()?.recordModelUsage(`${model.provider}/${model.id}`);
 
-		// Re-apply the current thinking level for the newly selected model
-		this.setThinkingLevel(this.thinkingLevel);
+		// Re-apply thinking for the newly selected model. Prefer the model's
+		// configured defaultLevel; otherwise preserve the current level.
+		this.setThinkingLevel(model.thinking?.defaultLevel ?? this.thinkingLevel);
 		await this.#syncEditToolModeAfterModelChange(previousEditMode);
 	}
 
@@ -3577,8 +3650,9 @@ export class AgentSession {
 		this.sessionManager.appendModelChange(`${model.provider}/${model.id}`, "temporary");
 		this.settings.getStorage()?.recordModelUsage(`${model.provider}/${model.id}`);
 
-		// Apply explicit thinking level, or re-clamp current level to new model's capabilities
-		this.setThinkingLevel(thinkingLevel ?? this.thinkingLevel);
+		// Apply explicit thinking level if given; otherwise prefer the model's
+		// configured defaultLevel; otherwise re-clamp the current level.
+		this.setThinkingLevel(thinkingLevel ?? model.thinking?.defaultLevel ?? this.thinkingLevel);
 		await this.#syncEditToolModeAfterModelChange(previousEditMode);
 	}
 
@@ -3876,9 +3950,13 @@ export class AgentSession {
 	 * @param options Optional callbacks for completion/error handling
 	 */
 	async compact(customInstructions?: string, options?: CompactOptions): Promise<CompactionResult> {
+		if (this.#compactionAbortController) {
+			throw new Error("Compaction already in progress");
+		}
 		this.#disconnectFromAgent();
 		await this.abort();
-		this.#compactionAbortController = new AbortController();
+		const compactionAbortController = new AbortController();
+		this.#compactionAbortController = compactionAbortController;
 
 		try {
 			if (!this.model) {
@@ -3916,7 +3994,7 @@ export class AgentSession {
 					preparation,
 					branchEntries: pathEntries,
 					customInstructions,
-					signal: this.#compactionAbortController.signal,
+					signal: compactionAbortController.signal,
 				})) as SessionBeforeCompactResult | undefined;
 
 				if (result?.cancel) {
@@ -3963,7 +4041,7 @@ export class AgentSession {
 					compactionModel,
 					apiKey,
 					customInstructions,
-					this.#compactionAbortController.signal,
+					compactionAbortController.signal,
 					{ promptOverride: hookPrompt, extraContext: hookContext, remoteInstructions: this.#baseSystemPrompt },
 				);
 				summary = result.summary;
@@ -3974,7 +4052,7 @@ export class AgentSession {
 				preserveData = { ...(preserveData ?? {}), ...(result.preserveData ?? {}) };
 			}
 
-			if (this.#compactionAbortController.signal.aborted) {
+			if (compactionAbortController.signal.aborted) {
 				throw new Error("Compaction cancelled");
 			}
 
@@ -4021,7 +4099,9 @@ export class AgentSession {
 			options?.onError?.(err);
 			throw error;
 		} finally {
-			this.#compactionAbortController = undefined;
+			if (this.#compactionAbortController === compactionAbortController) {
+				this.#compactionAbortController = undefined;
+			}
 			this.#reconnectToAgent();
 		}
 	}
@@ -5263,9 +5343,12 @@ export class AgentSession {
 
 	#isTransientTransportErrorMessage(errorMessage: string): boolean {
 		// Match: overloaded_error, provider returned error, rate limit, 429, 500, 502, 503, 504,
-		// service unavailable, network/connection errors, fetch failed, terminated, retry delay exceeded
-		return /overloaded|provider.?returned.?error|rate.?limit|too many requests|429|500|502|503|504|service.?unavailable|server.?error|internal.?error|network.?error|connection.?error|connection.?refused|other side closed|fetch failed|upstream.?connect|reset before headers|socket hang up|timed? out|timeout|terminated|retry delay|stream stall/i.test(
-			errorMessage,
+		// service unavailable, network/connection/socket errors, fetch failed, terminated, retry delay exceeded
+		return (
+			isUnexpectedSocketCloseMessage(errorMessage) ||
+			/overloaded|provider.?returned.?error|rate.?limit|too many requests|429|500|502|503|504|service.?unavailable|server.?error|internal.?error|network.?error|connection.?error|connection.?refused|other side closed|fetch failed|upstream.?connect|reset before headers|socket hang up|timed? out|timeout|terminated|retry delay|stream stall/i.test(
+				errorMessage,
+			)
 		);
 	}
 
@@ -5608,15 +5691,16 @@ export class AgentSession {
 			this.agent.replaceMessages(messages.slice(0, -1));
 		}
 
-		// Wait with exponential backoff (abortable)
-		// Properly abort and null existing controller before replacing
-		if (this.#retryAbortController) {
-			this.#retryAbortController.abort();
-		}
-		this.#retryAbortController = new AbortController();
+		// Wait with exponential backoff (abortable).
+		const retryAbortController = new AbortController();
+		this.#retryAbortController?.abort();
+		this.#retryAbortController = retryAbortController;
 		try {
-			await abortableSleep(delayMs, this.#retryAbortController.signal);
+			await abortableSleep(delayMs, retryAbortController.signal);
 		} catch {
+			if (this.#retryAbortController !== retryAbortController) {
+				return false;
+			}
 			// Aborted during sleep - emit end event so UI can clean up
 			const attempt = this.#retryAttempt;
 			this.#retryAttempt = 0;
@@ -5630,7 +5714,9 @@ export class AgentSession {
 			this.#resolveRetry();
 			return false;
 		}
-		this.#retryAbortController = undefined;
+		if (this.#retryAbortController === retryAbortController) {
+			this.#retryAbortController = undefined;
+		}
 
 		// Retry via continue() outside the agent_end event callback chain.
 		this.#scheduleAgentContinue({ delayMs: 1, generation });
@@ -5722,12 +5808,13 @@ export class AgentSession {
 			}
 		}
 
-		this.#bashAbortController = new AbortController();
+		const abortController = new AbortController();
+		this.#bashAbortControllers.add(abortController);
 
 		try {
 			const result = await executeBashCommand(command, {
 				onChunk,
-				signal: this.#bashAbortController.signal,
+				signal: abortController.signal,
 				sessionKey: this.sessionId,
 				timeout: clampTimeout("bash") * 1000,
 				onMinimizedSave: originalText => this.#saveBashOriginalArtifact(originalText),
@@ -5736,7 +5823,7 @@ export class AgentSession {
 			this.recordBashResult(command, result, options);
 			return result;
 		} finally {
-			this.#bashAbortController = undefined;
+			this.#bashAbortControllers.delete(abortController);
 		}
 	}
 
@@ -5775,12 +5862,14 @@ export class AgentSession {
 	 * Cancel running bash command.
 	 */
 	abortBash(): void {
-		this.#bashAbortController?.abort();
+		for (const abortController of this.#bashAbortControllers) {
+			abortController.abort();
+		}
 	}
 
 	/** Whether a bash command is currently running */
 	get isBashRunning(): boolean {
-		return this.#bashAbortController !== undefined;
+		return this.#bashAbortControllers.size > 0;
 	}
 
 	/** Whether there are pending bash messages waiting to be flushed */
@@ -6518,6 +6607,8 @@ export class AgentSession {
 		cancelled: boolean;
 		aborted?: boolean;
 		summaryEntry?: BranchSummaryEntry;
+		/** Raw session context built during navigation — pass to renderInitialMessages to skip a second O(N) walk. */
+		sessionContext?: SessionContext;
 	}> {
 		const oldLeafId = this.sessionManager.getLeafId();
 
@@ -6647,15 +6738,20 @@ export class AgentSession {
 			this.sessionManager.branch(newLeafId);
 		}
 
-		// Update agent state
-		const sessionContext = this.buildDisplaySessionContext();
-		await this.#restoreMCPSelectionsForSessionContext(sessionContext);
-		this.agent.replaceMessages(sessionContext.messages);
+		// Update agent state — build display context to populate agent messages.
+		const stateContext = this.sessionManager.buildSessionContext();
+		const displayContext = deobfuscateSessionContext(stateContext, this.#obfuscator);
+		await this.#restoreMCPSelectionsForSessionContext(displayContext);
+		this.agent.replaceMessages(displayContext.messages);
 		this.#syncTodoPhasesFromBranch();
 		this.#closeCodexProviderSessionsForHistoryRewrite();
 
-		// Emit session_tree event
-		if (this.#extensionRunner) {
+		this.#branchSummaryAbortController = undefined;
+
+		// Emit session_tree event; only handlers can mutate session entries, so skip
+		// the emit and the context rebuild when no handlers are registered (mirrors
+		// the session_before_tree guard above).
+		if (this.#extensionRunner?.hasHandlers("session_tree")) {
 			await this.#extensionRunner.emit({
 				type: "session_tree",
 				newLeafId: this.sessionManager.getLeafId(),
@@ -6663,10 +6759,10 @@ export class AgentSession {
 				summaryEntry,
 				fromExtension: summaryText ? fromExtension : undefined,
 			});
+			const rawContext = this.sessionManager.buildSessionContext();
+			return { editorText, cancelled: false, summaryEntry, sessionContext: rawContext };
 		}
-
-		this.#branchSummaryAbortController = undefined;
-		return { editorText, cancelled: false, summaryEntry };
+		return { editorText, cancelled: false, summaryEntry, sessionContext: stateContext };
 	}
 
 	/**