@oh-my-pi/pi-ai 13.3.2 → 13.3.3

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-ai",
-	"version": "13.3.2",
+	"version": "13.3.3",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"homepage": "https://github.com/can1357/oh-my-pi",
 	"author": "Can Boluk",
@@ -38,19 +38,19 @@
 	},
 	"dependencies": {
 		"@anthropic-ai/sdk": "^0.78",
-		"@aws-sdk/client-bedrock-runtime": "^3.995",
+		"@aws-sdk/client-bedrock-runtime": "^3.998",
 		"@bufbuild/protobuf": "^2.11",
 		"@connectrpc/connect": "^2.1",
 		"@connectrpc/connect-node": "^2.1",
-		"@google/genai": "^1.42",
+		"@google/genai": "^1.43",
 		"@mistralai/mistralai": "^1.14",
-		"@oh-my-pi/pi-utils": "13.3.2",
+		"@oh-my-pi/pi-utils": "13.3.3",
 		"@sinclair/typebox": "^0.34",
 		"@smithy/node-http-handler": "^4.4",
 		"ajv": "^8.18",
 		"ajv-formats": "^3.0",
 		"chalk": "^5.6",
-		"openai": "^6.22",
+		"openai": "^6.25",
 		"partial-json": "^0.1",
 		"zod": "^4.3",
 		"zod-to-json-schema": "^3.25"

package/src/provider-models/openai-compat.ts

@@ -1670,6 +1670,7 @@ const MODELS_DEV_PROVIDER_DESCRIPTORS_CODING_PLANS: readonly ModelsDevProviderDe
 	// --- MiniMax Coding Plan ---
 	openAiCompletionsDescriptor("minimax-coding-plan", "minimax-code", "https://api.minimax.io/v1", {
 		compat: {
+			supportsStore: false,
 			supportsDeveloperRole: false,
 			thinkingFormat: "zai",
 			reasoningContentField: "reasoning_content",
@@ -1677,6 +1678,7 @@ const MODELS_DEV_PROVIDER_DESCRIPTORS_CODING_PLANS: readonly ModelsDevProviderDe
 	}),
 	openAiCompletionsDescriptor("minimax-cn-coding-plan", "minimax-code-cn", "https://api.minimaxi.com/v1", {
 		compat: {
+			supportsStore: false,
 			supportsDeveloperRole: false,
 			thinkingFormat: "zai",
 			reasoningContentField: "reasoning_content",

package/src/providers/anthropic.ts

@@ -296,12 +296,18 @@ const PROVIDER_BASE_DELAY_MS = 2000;
  * Includes malformed JSON stream-envelope parse errors seen from some
  * Anthropic-compatible proxy endpoints.
  */
+/** Transient stream corruption errors where the response was truncated mid-JSON. */
+function isTransientStreamParseError(error: unknown): boolean {
+	if (!(error instanceof Error)) return false;
+	return /json parse error|unterminated string|unexpected end of json input/i.test(error.message);
+}
+
 export function isProviderRetryableError(error: unknown): boolean {
 	if (!(error instanceof Error)) return false;
 	const msg = error.message;
 	return (
 		/rate.?limit|too many requests|overloaded|service.?unavailable|1302/i.test(msg) ||
-		/json parse error|unterminated string|unexpected end of json input/i.test(msg)
+		isTransientStreamParseError(error)
 	);
 }
 
@@ -371,9 +377,9 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 			type Block = (ThinkingContent | TextContent | (ToolCall & { partialJson: string })) & { index: number };
 			const blocks = output.content as Block[];
 			stream.push({ type: "start", partial: output });
-			// Retry loop for rate-limit errors from proxies (e.g. z.ai) that the SDK doesn't handle.
-			// These errors surface when iterating the stream, so we retry the full stream creation.
-			// Only retry if no content blocks have been emitted yet (safe to restart).
+			// Retry loop for transient errors from the stream.
+			// Rate-limit/overload: only before content starts (safe to restart).
+			// Truncated JSON: also after content starts (partial response is unusable).
 			let providerRetryAttempt = 0;
 			let started = false;
 			do {
@@ -542,12 +548,15 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 					}
 					break; // Stream completed successfully
 				} catch (streamError) {
-					// Only retry if: not aborted, no content emitted yet, retries left, and error is retryable
+					// Transient stream parse errors (truncated JSON) are retryable even after content
+					// has started streaming, since the partial response is unusable anyway.
+					// Rate-limit/overload errors are only retried before content starts.
+					const isTransient = isTransientStreamParseError(streamError);
 					if (
 						options?.signal?.aborted ||
-						firstTokenTime !== undefined ||
 						providerRetryAttempt >= PROVIDER_MAX_RETRIES ||
-						!isProviderRetryableError(streamError)
+						(!isTransient && firstTokenTime !== undefined) ||
+						(!isTransient && !isProviderRetryableError(streamError))
 					) {
 						throw streamError;
 					}
@@ -557,6 +566,8 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 					// Reset output state for clean retry
 					output.content.length = 0;
 					output.stopReason = "stop";
+					firstTokenTime = undefined;
+					started = false;
 				}
 			} while (!started);
 

package/src/providers/cursor.ts

@@ -37,6 +37,7 @@ import {
 	AssistantMessageSchema,
 	BackgroundShellSpawnResultSchema,
 	ClientHeartbeatSchema,
+	ComputerUseResultSchema,
 	ConversationActionSchema,
 	type ConversationStateStructure,
 	ConversationStateStructureSchema,
@@ -70,6 +71,7 @@ import {
 	GrepUnionResultSchema,
 	KvClientMessageSchema,
 	type KvServerMessage,
+	ListMcpResourcesExecResultSchema,
 	type LsDirectoryTreeNode,
 	type LsDirectoryTreeNode_File,
 	LsDirectoryTreeNode_FileSchema,
@@ -88,9 +90,11 @@ import {
 	McpToolResultContentItemSchema,
 	ModelDetailsSchema,
 	ReadErrorSchema,
+	ReadMcpResourceExecResultSchema,
 	ReadRejectedSchema,
 	ReadResultSchema,
 	ReadSuccessSchema,
+	RecordScreenResultSchema,
 	RequestContextResultSchema,
 	RequestContextSchema,
 	RequestContextSuccessSchema,
@@ -965,6 +969,26 @@ async function handleExecServerMessage(
 			sendExecClientMessage(h2Request, execMsg, "mcpResult", execResult);
 			return;
 		}
+		case "listMcpResourcesExecArgs": {
+			const execResult = create(ListMcpResourcesExecResultSchema, {});
+			sendExecClientMessage(h2Request, execMsg, "listMcpResourcesExecResult", execResult);
+			return;
+		}
+		case "readMcpResourceExecArgs": {
+			const execResult = create(ReadMcpResourceExecResultSchema, {});
+			sendExecClientMessage(h2Request, execMsg, "readMcpResourceExecResult", execResult);
+			return;
+		}
+		case "recordScreenArgs": {
+			const execResult = create(RecordScreenResultSchema, {});
+			sendExecClientMessage(h2Request, execMsg, "recordScreenResult", execResult);
+			return;
+		}
+		case "computerUseArgs": {
+			const execResult = create(ComputerUseResultSchema, {});
+			sendExecClientMessage(h2Request, execMsg, "computerUseResult", execResult);
+			return;
+		}
 		default:
 			log("warn", "unhandledExecMessage", { execCase });
 	}