@oh-my-pi/pi-ai 17.1.2 → 17.1.3

package/CHANGELOG.md

@@ -2,6 +2,13 @@
 
 ## [Unreleased]
 
+## [17.1.3] - 2026-07-24
+
+### Fixed
+
+- Fixed Cursor sessions exposing `ast_edit` (and other staged-preview `xd://` devices) without a reachable resolver: the built-in `write` tool — which carries the `xd://resolve` / `xd://reject` transport that finalizes a staged preview — was filtered out of Cursor's forwarded catalog, so previews could never be resolved and the session aborted after three forced `write` turns. `write` is now re-included in the forwarded catalog whenever pi-agent devices are advertised ([#6536](https://github.com/can1357/oh-my-pi/issues/6536)).
+- Fixed OpenAI Responses and chat-completions streams honoring per-model first-event watchdog policy, allowing local llama.cpp-style backends to process arbitrarily large prompts without a premature client cancellation ([#6524](https://github.com/can1357/oh-my-pi/issues/6524)).
+
 ## [17.1.2] - 2026-07-24
 
 ### Added

package/dist/types/providers/cursor.d.ts

@@ -2,7 +2,7 @@ import http2 from "node:http2";
 import { type JsonValue } from "@bufbuild/protobuf";
 import type { McpToolDefinition } from "@oh-my-pi/pi-catalog/discovery/cursor-gen/agent_pb";
 import { type AgentServerMessage, type ConversationStateStructure } from "@oh-my-pi/pi-catalog/discovery/cursor-gen/agent_pb";
-import type { AssistantMessage, CursorExecHandlerResult, CursorExecHandlers, CursorToolResultHandler, Message, StreamFunction, StreamOptions, TextContent, ThinkingContent, ToolCall, ToolResultMessage } from "../types.js";
+import type { AssistantMessage, CursorExecHandlerResult, CursorExecHandlers, CursorToolResultHandler, Message, StreamFunction, StreamOptions, TextContent, ThinkingContent, Tool, ToolCall, ToolResultMessage } from "../types.js";
 import { kCursorExecResolved, kStreamingBlockIndex, kStreamingBlockKind, kStreamingLastParseLen, kStreamingPartialJson } from "../utils/block-symbols.js";
 import { AssistantMessageEventStream } from "../utils/event-stream.js";
 export declare const CURSOR_API_URL = "https://api2.cursor.sh";
@@ -119,6 +119,7 @@ export declare function mergeCursorMcpToolCallArgs(streamed: Record<string, unkn
 export declare function synthesizeCursorExecToolCall(output: AssistantMessage, stream: AssistantMessageEventStream, state: BlockState, toolCallId: string, toolName: string, args: Record<string, unknown>): void;
 /** Exported for tests: drives one Cursor interaction update through the streaming state machine. */
 export declare function processInteractionUpdate(update: any, output: AssistantMessage, stream: AssistantMessageEventStream, state: BlockState, usageState: UsageState): void;
+export declare function buildMcpToolDefinitions(tools: Tool[] | undefined): McpToolDefinition[];
 /**
  * Build `ConversationStateStructure.rootPromptMessagesJson` blob IDs for the
  * system prompt plus prior conversation history, as JSON blobs matching

package/dist/types/utils/idle-iterator.d.ts

@@ -39,11 +39,13 @@ export declare function getStreamFirstEventTimeoutMs(idleTimeoutMs?: number, fal
  * `"0"` disable) wins outright. Otherwise the resolved idle (caller-supplied
  * `idleTimeoutMs` — which itself already encompasses per-call
  * `streamIdleTimeoutMs` or `PI_OPENAI_STREAM_IDLE_TIMEOUT_MS` resolved
- * upstream) floors the first-event budget so slow local OpenAI-compatible
- * servers are not undercut by a shorter `PI_STREAM_FIRST_EVENT_TIMEOUT_MS`
- * or the global default during prompt processing.
+ * upstream) floors the first-event budget so slow OpenAI-compatible servers
+ * are not undercut by a shorter `PI_STREAM_FIRST_EVENT_TIMEOUT_MS` or the
+ * global default during prompt processing. A zero per-provider fallback
+ * disables the first-event watchdog unless an environment override is set.
  *
- * Returns `undefined` when an explicit env knob disables the watchdog.
+ * Returns `0` when an explicit env knob or per-provider fallback disables the
+ * watchdog, preserving the sentinel through iterator timeout resolution.
  */
 export declare function getOpenAIStreamFirstEventTimeoutMs(idleTimeoutMs?: number, fallbackMs?: number): number | undefined;
 /**

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-ai",
-	"version": "17.1.2",
+	"version": "17.1.3",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"homepage": "https://omp.sh",
 	"author": "Can Boluk",
@@ -38,9 +38,9 @@
 	},
 	"dependencies": {
 		"@bufbuild/protobuf": "^2.12.1",
-		"@oh-my-pi/pi-catalog": "17.1.2",
-		"@oh-my-pi/pi-utils": "17.1.2",
-		"@oh-my-pi/pi-wire": "17.1.2",
+		"@oh-my-pi/pi-catalog": "17.1.3",
+		"@oh-my-pi/pi-utils": "17.1.3",
+		"@oh-my-pi/pi-wire": "17.1.3",
 		"arktype": "2.2.3",
 		"zod": "^4"
 	},

package/src/providers/cursor.ts

@@ -2401,7 +2401,7 @@ function readCursorBlob(blobStore: Map<string, Uint8Array>, blobId: Uint8Array):
 
 const CURSOR_NATIVE_TOOL_NAMES = new Set(["bash", "read", "write", "delete", "ls", "grep", "lsp", "todo"]);
 
-function buildMcpToolDefinitions(tools: Tool[] | undefined): McpToolDefinition[] {
+export function buildMcpToolDefinitions(tools: Tool[] | undefined): McpToolDefinition[] {
 	if (!tools || tools.length === 0) {
 		return [];
 	}
@@ -2411,7 +2411,16 @@ function buildMcpToolDefinitions(tools: Tool[] | undefined): McpToolDefinition[]
 		return [];
 	}
 
-	return advertisedTools.map(tool => {
+	// The `write` tool doubles as the xd:// transport: forwarded devices such as
+	// `ast_edit` stage previews finalized only by writing a reason to xd://resolve
+	// or xd://reject. Cursor's native catalog may expose no write path, so
+	// re-include the built-in `write` (dropped as native above) whenever pi-agent
+	// devices are advertised — otherwise a staged preview can never be resolved
+	// and the SoftToolRequirement('write') escalation aborts the turn.
+	const writeTool = tools.find(tool => tool.name === "write");
+	const forwarded = writeTool ? [...advertisedTools, writeTool] : advertisedTools;
+
+	return forwarded.map(tool => {
 		const jsonSchema = toolWireSchema(tool);
 		const schemaValue: JsonValue =
 			jsonSchema && typeof jsonSchema === "object"

package/src/providers/openai-completions.ts

@@ -630,7 +630,8 @@ const streamOpenAICompletionsOnce = (
 			const idleTimeoutFallbackMs = model.compat.streamIdleTimeoutMs;
 			const idleTimeoutMs = options?.streamIdleTimeoutMs ?? getOpenAIStreamIdleTimeoutMs(idleTimeoutFallbackMs);
 			const firstEventTimeoutMs =
-				options?.streamFirstEventTimeoutMs ?? getOpenAIStreamFirstEventTimeoutMs(idleTimeoutMs);
+				options?.streamFirstEventTimeoutMs ??
+				getOpenAIStreamFirstEventTimeoutMs(idleTimeoutMs, model.compat.streamFirstEventTimeoutMs);
 			const requestTimeoutMs =
 				firstEventTimeoutMs !== undefined && firstEventTimeoutMs > 0 ? firstEventTimeoutMs : undefined;
 			const { copilotPremiumRequests, baseUrl, headers, query, requestHeaders } = createRequestSetup(

package/src/providers/openai-responses.ts

@@ -495,7 +495,8 @@ const streamOpenAIResponsesOnce = (
 			const idleTimeoutMs =
 				options?.streamIdleTimeoutMs ?? getOpenAIStreamIdleTimeoutMs(model.compat.streamIdleTimeoutMs);
 			const firstEventTimeoutMs =
-				options?.streamFirstEventTimeoutMs ?? getOpenAIStreamFirstEventTimeoutMs(idleTimeoutMs);
+				options?.streamFirstEventTimeoutMs ??
+				getOpenAIStreamFirstEventTimeoutMs(idleTimeoutMs, model.compat.streamFirstEventTimeoutMs);
 			const requestTimeoutMs =
 				firstEventTimeoutMs !== undefined && firstEventTimeoutMs > 0 ? firstEventTimeoutMs : undefined;
 			const requestUrl = `${resolvedBaseUrl}/responses`;

package/src/utils/idle-iterator.ts

@@ -68,11 +68,13 @@ export function getStreamFirstEventTimeoutMs(
  * `"0"` disable) wins outright. Otherwise the resolved idle (caller-supplied
  * `idleTimeoutMs` — which itself already encompasses per-call
  * `streamIdleTimeoutMs` or `PI_OPENAI_STREAM_IDLE_TIMEOUT_MS` resolved
- * upstream) floors the first-event budget so slow local OpenAI-compatible
- * servers are not undercut by a shorter `PI_STREAM_FIRST_EVENT_TIMEOUT_MS`
- * or the global default during prompt processing.
+ * upstream) floors the first-event budget so slow OpenAI-compatible servers
+ * are not undercut by a shorter `PI_STREAM_FIRST_EVENT_TIMEOUT_MS` or the
+ * global default during prompt processing. A zero per-provider fallback
+ * disables the first-event watchdog unless an environment override is set.
  *
- * Returns `undefined` when an explicit env knob disables the watchdog.
+ * Returns `0` when an explicit env knob or per-provider fallback disables the
+ * watchdog, preserving the sentinel through iterator timeout resolution.
  */
 export function getOpenAIStreamFirstEventTimeoutMs(
 	idleTimeoutMs?: number,
@@ -80,10 +82,10 @@ export function getOpenAIStreamFirstEventTimeoutMs(
 ): number | undefined {
 	const openAIFirstEventRaw = $env.PI_OPENAI_STREAM_FIRST_EVENT_TIMEOUT_MS;
 	if (openAIFirstEventRaw !== undefined) {
-		return normalizeIdleTimeoutMs(openAIFirstEventRaw, fallbackMs);
+		return normalizeIdleTimeoutMs(openAIFirstEventRaw, fallbackMs) ?? 0;
 	}
 	const base = normalizeIdleTimeoutMs($env.PI_STREAM_FIRST_EVENT_TIMEOUT_MS, fallbackMs);
-	if (base === undefined) return undefined;
+	if (base === undefined || base <= 0) return 0;
 	if (idleTimeoutMs === undefined || idleTimeoutMs <= 0) return base;
 	return Math.max(base, idleTimeoutMs);
 }