@oh-my-pi/pi-agent-core 16.2.2 → 16.2.4

package/CHANGELOG.md

@@ -2,6 +2,29 @@
 
 ## [Unreleased]
 
+## [16.2.4] - 2026-06-28
+
+### Changed
+
+- Improved the reliability of remote compaction by introducing transient error retries, configurable timeouts, and immediate termination upon user-initiated aborts.
+
+### Fixed
+
+- Fixed an issue where assistant responses and encrypted reasoning could be lost during local history trimming prior to remote compaction.
+- Fixed type compatibility for hosts with title audit entries by adding support for `title_change` session metadata.
+- Fixed an issue where transient stream read failures after a completed tool call were treated as terminal errors, allowing the agent to successfully execute the tool and continue the turn.
+
+## [16.2.3] - 2026-06-28
+
+### Changed
+
+- Enabled V2 streaming remote compaction by default for compatible AI and OpenAI-compatible models, which forwards full conversation history to the provider and supports session routing, prompt caching, provider-native tool history replay, transient error retries, and configurable timeouts.
+
+### Fixed
+
+- Fixed an issue where assistant responses and encrypted reasoning could be lost during local history trimming.
+- Added `title_change` session metadata to the compaction entry type union to maintain type compatibility for hosts with title audit entries.
+
 ## [16.2.2] - 2026-06-27
 
 ### Added

package/dist/types/compaction/compaction-v2-streaming.d.ts

@@ -0,0 +1,82 @@
+/**
+ * Remote Compaction V2: streaming Responses compaction.
+ *
+ * Mirrors Codex `core/src/compact_remote_v2.rs`: append a `compaction_trigger`
+ * input item to the normal Responses stream, require exactly one streamed
+ * compaction output item, then install retained real user messages plus that
+ * compaction item as replacement history.
+ */
+import type { FetchImpl, Model } from "@oh-my-pi/pi-ai";
+/** Retained-message budget Codex uses after streamed V2 compaction. */
+export declare const V2_RETAINED_MESSAGE_TOKEN_BUDGET = 64000;
+/** Max retries for V2 streaming compaction on transient stream errors. */
+export declare const V2_COMPACTION_MAX_RETRIES = 2;
+/** Timeout for V2 streaming compaction (3 minutes, same as V1). */
+export declare const V2_COMPACTION_TIMEOUT_MS = 180000;
+/** Token usage reported by the streamed V2 Responses completion. */
+export interface CompactionV2Usage {
+    inputTokens: number;
+    outputTokens: number;
+    totalTokens: number;
+    cachedInputTokens?: number;
+    reasoningOutputTokens?: number;
+}
+/** Request body fields needed for Responses-stream V2 compaction. */
+export interface CompactionV2Request {
+    model: string;
+    input: unknown[];
+    instructions: string;
+    retainedMessageBudget: number;
+    tools?: unknown[];
+    /** Responses reasoning param (effort + summary), matching a normal turn; omitted for non-reasoning models. */
+    reasoning?: {
+        effort: string;
+        summary: string;
+    };
+    sessionId?: string;
+    promptCacheKey?: string;
+}
+/** Response collected from the V2 stream and converted into replacement history. */
+export interface CompactionV2Response {
+    compactionItem: Record<string, unknown>;
+    replacementHistory: Array<Record<string, unknown>>;
+    usedTokens: number;
+    usage?: CompactionV2Usage;
+    retainedImageCount: number;
+}
+/** Resolve the streaming Responses endpoint for a V2-capable model. */
+export declare function getCompactionV2Endpoint(model: Model): string | undefined;
+/** Check whether a model can use streaming V2 compaction. */
+export declare function shouldUseCompactionV2Streaming(model: Model): model is Model<"openai-responses" | "azure-openai-responses" | "openai-codex-responses">;
+/** Clamp the retained-message budget to Codex's known-safe 64K ceiling. */
+export declare function resolveCompactionV2RetainedMessageBudget(value: number | undefined): number;
+/** Build a V2 streaming compaction request from Responses-native history. */
+export declare function buildCompactionV2Request(model: Model, input: unknown[], instructions: string, options?: {
+    tools?: unknown[];
+    reasoning?: {
+        effort: string;
+        summary: string;
+    };
+    sessionId?: string;
+    promptCacheKey?: string;
+    retainedMessageBudget?: number;
+}): CompactionV2Request;
+/** Request V2 compaction over the normal OpenAI Responses streaming endpoint. */
+export declare function requestCompactionV2Streaming(model: Model, apiKey: string, request: CompactionV2Request, signal?: AbortSignal, options?: {
+    fetch?: FetchImpl;
+    timeoutMs?: number;
+    retryWait?: (delayMs: number, signal?: AbortSignal) => Promise<void>;
+}): Promise<CompactionV2Response>;
+/** Build Codex-style V2 replacement history from prompt input plus compaction output. */
+export declare function buildCompactionV2ReplacementHistory(input: unknown[], compactionItem: Record<string, unknown>, retainedMessageBudget?: number): {
+    replacementHistory: Array<Record<string, unknown>>;
+    retainedImageCount: number;
+};
+/** Store V2 replacement history in the OpenAI remote-compaction preserve slot. */
+export declare function storeCompactionV2PreserveData(response: CompactionV2Response, model: Model): Record<string, unknown>;
+/** Retrieve preserved OpenAI replacement history that V2 can extend. */
+export declare function getCompactionV2PreserveData(preserveData: Record<string, unknown> | undefined): {
+    provider: string;
+    replacementHistory: Array<Record<string, unknown>>;
+    usedTokens: number;
+} | undefined;

package/dist/types/compaction/compaction.d.ts

@@ -39,6 +39,8 @@ export interface CompactionSettings {
     autoContinue?: boolean;
     remoteEnabled?: boolean;
     remoteEndpoint?: string;
+    remoteStreamingV2Enabled?: boolean;
+    v2RetainedMessageBudget?: number;
 }
 export declare const DEFAULT_COMPACTION_SETTINGS: CompactionSettings;
 /**
@@ -151,6 +153,12 @@ export interface SummaryOptions {
      * `resolveCompactionEffort` for the conversion contract.
      */
     thinkingLevel?: ThinkingLevel;
+    /** Session routing key for remote compaction transports with sticky provider sessions. */
+    sessionId?: string;
+    /** Prompt-cache key for remote compaction transports that support provider prefix caching. */
+    promptCacheKey?: string;
+    /** Provider-visible tools for remote compaction transports that replay native tool history. */
+    tools?: Tool[];
     /** Optional fetch implementation threaded into remote compaction calls. */
     fetch?: FetchImpl;
 }
@@ -229,7 +237,7 @@ export interface CompactionPreparation {
     /** Compaction settions from settings.jsonl	*/
     settings: CompactionSettings;
 }
-export declare function prepareCompaction(pathEntries: SessionEntry[], settings: CompactionSettings): CompactionPreparation | undefined;
+export declare function prepareCompaction(pathEntries: SessionEntry[], settings: CompactionSettings, compactionModels?: readonly Model[]): CompactionPreparation | undefined;
 /**
  * Generate summaries for compaction using prepared data.
  * Returns CompactionResult - SessionManager adds id/parentId when saving.

package/dist/types/compaction/entries.d.ts

@@ -66,6 +66,13 @@ export interface LabelEntry extends SessionEntryBase {
     targetId: string;
     label: string | undefined;
 }
+export interface TitleChangeEntry extends SessionEntryBase {
+    type: "title_change";
+    title: string;
+    previousTitle?: string;
+    source: "auto" | "user";
+    trigger?: string;
+}
 export interface TtsrInjectionEntry extends SessionEntryBase {
     type: "ttsr_injection";
     /** Names of rules that were injected */
@@ -96,7 +103,7 @@ export interface ModeChangeEntry extends SessionEntryBase {
 }
 export interface CustomCompactionSessionEntries {
 }
-export type SessionEntry = SessionMessageEntry | ThinkingLevelChangeEntry | ModelChangeEntry | ServiceTierChangeEntry | CompactionEntry | BranchSummaryEntry | CustomEntry | CustomMessageEntry | LabelEntry | TtsrInjectionEntry | MCPToolSelectionEntry | SessionInitEntry | ModeChangeEntry | CustomCompactionSessionEntries[keyof CustomCompactionSessionEntries];
+export type SessionEntry = SessionMessageEntry | ThinkingLevelChangeEntry | ModelChangeEntry | ServiceTierChangeEntry | CompactionEntry | BranchSummaryEntry | CustomEntry | CustomMessageEntry | LabelEntry | TitleChangeEntry | TtsrInjectionEntry | MCPToolSelectionEntry | SessionInitEntry | ModeChangeEntry | CustomCompactionSessionEntries[keyof CustomCompactionSessionEntries];
 export interface ReadonlySessionManager {
     getBranch(leafId?: string | null): SessionEntry[];
     getEntry(id: string): SessionEntry | undefined;

package/dist/types/compaction/openai.d.ts

@@ -1,9 +1,12 @@
 /**
  * Remote compaction utilities.
  *
- * Provider-side conversation summarization endpoints. Two flavors:
+ * Provider-side conversation summarization endpoints. Three flavors:
  *
- * - **OpenAI remote compaction** (`/responses/compact`): preserves encrypted
+ * - **OpenAI remote compaction V2** (Responses streaming): appends a
+ *   `compaction_trigger` input item to the normal stream and stores the returned
+ *   `compaction` item with retained real user messages in `preserveData`.
+ * - **OpenAI remote compaction V1** (`/responses/compact`): preserves encrypted
  *   reasoning across compactions by submitting the full responses-API native
  *   history and storing the returned `compaction` / `compaction_summary`
  *   item in `preserveData` so future turns can replay the encrypted state.
@@ -12,6 +15,7 @@
  *   with `{ summary, shortSummary? }`.
  */
 import type { FetchImpl, Message, Model } from "@oh-my-pi/pi-ai/types";
+export * from "./compaction-v2-streaming";
 export declare const OPENAI_REMOTE_COMPACTION_PRESERVE_KEY = "openaiRemoteCompaction";
 /**
  * Hard ceiling on remote compaction HTTP requests. Unlike every provider

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-agent-core",
-	"version": "16.2.2",
+	"version": "16.2.4",
 	"description": "General-purpose agent with transport abstraction, state management, and attachment support",
 	"homepage": "https://omp.sh",
 	"author": "Can Boluk",
@@ -35,12 +35,12 @@
 		"fmt": "biome format --write ."
 	},
 	"dependencies": {
-		"@oh-my-pi/pi-ai": "16.2.2",
-		"@oh-my-pi/pi-catalog": "16.2.2",
-		"@oh-my-pi/pi-natives": "16.2.2",
-		"@oh-my-pi/pi-utils": "16.2.2",
-		"@oh-my-pi/pi-wire": "16.2.2",
-		"@oh-my-pi/snapcompact": "16.2.2",
+		"@oh-my-pi/pi-ai": "16.2.4",
+		"@oh-my-pi/pi-catalog": "16.2.4",
+		"@oh-my-pi/pi-natives": "16.2.4",
+		"@oh-my-pi/pi-utils": "16.2.4",
+		"@oh-my-pi/pi-wire": "16.2.4",
+		"@oh-my-pi/snapcompact": "16.2.4",
 		"@opentelemetry/api": "^1.9.1"
 	},
 	"devDependencies": {

package/src/agent-loop.ts

@@ -1355,7 +1355,10 @@ async function streamAssistantResponse(
 
 					const event = next.value;
 					if (event.type === "done" || event.type === "error") {
-						let finalMessage = retainCompletedToolCalls(await response.result(), completedToolCallIds);
+						let finalMessage = recoverTransientErrorToolTurn(
+							retainCompletedToolCalls(await response.result(), completedToolCallIds),
+							context.tools ?? [],
+						);
 						if (harmonyMitigationEnabled) {
 							const detection = detectHarmonyLeakInAssistantMessage(finalMessage);
 							if (detection) {
@@ -1515,8 +1518,40 @@ function retainCompletedToolCalls(
 				: {
 						type: STREAM_INTERRUPTED_AFTER_CONTENT_STOP_DETAIL,
 						category: message.stopDetails?.type ?? null,
-						explanation: message.stopDetails?.explanation ?? null,
+						explanation: message.stopDetails?.explanation ?? message.errorMessage ?? null,
+					},
+	};
+}
+
+function recoverTransientErrorToolTurn(
+	message: AssistantMessage,
+	availableTools: ReadonlyArray<Pick<AgentTool, "name" | "customWireName">>,
+): AssistantMessage {
+	if (message.stopReason !== "error") return message;
+	const toolCalls = message.content.filter(block => block.type === "toolCall");
+	if (toolCalls.length === 0) return message;
+	const availableToolNames = new Set<string>();
+	for (const tool of availableTools) {
+		availableToolNames.add(tool.name);
+		if (tool.customWireName !== undefined) availableToolNames.add(tool.customWireName);
+	}
+	if (!toolCalls.every(toolCall => availableToolNames.has(toolCall.name))) return message;
+	if (!AIError.isStreamReadErrorText(`${message.errorMessage ?? ""}\n${message.stopDetails?.explanation ?? ""}`))
+		return message;
+	return {
+		...message,
+		stopReason: "toolUse",
+		stopDetails:
+			message.stopDetails?.type === STREAM_INTERRUPTED_AFTER_CONTENT_STOP_DETAIL
+				? message.stopDetails
+				: {
+						type: STREAM_INTERRUPTED_AFTER_CONTENT_STOP_DETAIL,
+						category: message.stopDetails?.type ?? null,
+						explanation: message.stopDetails?.explanation ?? message.errorMessage ?? null,
 					},
+		errorMessage: undefined,
+		errorId: undefined,
+		errorStatus: undefined,
 	};
 }