@sentry/junior 0.27.2 → 0.29.0

package/dist/app.js

@@ -6,11 +6,17 @@ import {
   parseSkillInvocation
 } from "./chunk-ICIRAL6Y.js";
 import {
+  GEN_AI_PROVIDER_NAME,
+  MISSING_GATEWAY_CREDENTIALS_ERROR,
   SANDBOX_DATA_ROOT,
   SANDBOX_SKILLS_ROOT,
   SANDBOX_WORKSPACE_ROOT,
   botConfig,
   buildNonInteractiveShellScript,
+  completeObject,
+  completeText,
+  getGatewayApiKey,
+  getPiGatewayApiKeyOverride,
   getRuntimeDependencyProfileHash,
   getRuntimeMetadata,
   getSlackBotToken,
@@ -20,19 +26,18 @@ import {
   getStateAdapter,
   getVercelSandboxCredentials,
   isSnapshotMissingError,
+  resolveGatewayModel,
   resolveRuntimeDependencySnapshot,
   runNonInteractiveCommand,
   sandboxSkillDir,
-  sandboxSkillFile,
-  toOptionalTrimmed
-} from "./chunk-4PVJHUEV.js";
+  sandboxSkillFile
+} from "./chunk-LEYD42MR.js";
 import {
   CredentialUnavailableError,
   buildOAuthTokenRequest,
   createChatSdkLogger,
   createPluginBroker,
   createRequestContext,
-  extractGenAiUsageAttributes,
   extractGenAiUsageSummary,
   getActiveTraceId,
   getPluginDefinition,
@@ -2066,228 +2071,6 @@ function getTurnUserReplyAttachmentContext(message) {
   };
 }
 
-// src/chat/pi/client.ts
-import {
-  completeSimple,
-  getEnvApiKey,
-  getModels,
-  registerApiProvider
-} from "@mariozechner/pi-ai";
-import {
-  streamAnthropic,
-  streamSimpleAnthropic
-} from "@mariozechner/pi-ai/anthropic";
-registerApiProvider({
-  api: "anthropic-messages",
-  stream: streamAnthropic,
-  streamSimple: streamSimpleAnthropic
-});
-var GATEWAY_PROVIDER = "vercel-ai-gateway";
-var GEN_AI_PROVIDER_NAME = GATEWAY_PROVIDER;
-var GEN_AI_OPERATION_CHAT = "chat";
-var MISSING_GATEWAY_CREDENTIALS_ERROR = "Missing AI gateway credentials (AI_GATEWAY_API_KEY or VERCEL_OIDC_TOKEN)";
-function getGatewayApiKey() {
-  return toOptionalTrimmed(getEnvApiKey("vercel-ai-gateway")) ?? toOptionalTrimmed(process.env.VERCEL_OIDC_TOKEN);
-}
-function getPiGatewayApiKeyOverride() {
-  return toOptionalTrimmed(process.env.VERCEL_OIDC_TOKEN);
-}
-function extractText(message) {
-  return (message.content ?? []).filter((part) => part.type === "text" && typeof part.text === "string").map((part) => part.text ?? "").join("").trim();
-}
-function parseJsonCandidate(text) {
-  const trimmed = text.trim();
-  if (!trimmed) return void 0;
-  try {
-    return JSON.parse(trimmed);
-  } catch {
-    const fencedBlocks = [
-      ...trimmed.matchAll(/```(?:json)?\s*([\s\S]*?)\s*```/gi)
-    ];
-    for (const block of fencedBlocks) {
-      try {
-        return JSON.parse(block[1]);
-      } catch {
-      }
-    }
-    const openBraceIndex = trimmed.indexOf("{");
-    if (openBraceIndex >= 0) {
-      let depth = 0;
-      let inString = false;
-      let escaped = false;
-      for (let index = openBraceIndex; index < trimmed.length; index += 1) {
-        const char = trimmed[index];
-        if (inString) {
-          if (escaped) {
-            escaped = false;
-            continue;
-          }
-          if (char === "\\") {
-            escaped = true;
-            continue;
-          }
-          if (char === '"') {
-            inString = false;
-          }
-          continue;
-        }
-        if (char === '"') {
-          inString = true;
-          continue;
-        }
-        if (char === "{") {
-          depth += 1;
-          continue;
-        }
-        if (char === "}") {
-          depth -= 1;
-          if (depth === 0) {
-            const slice = trimmed.slice(openBraceIndex, index + 1);
-            try {
-              return JSON.parse(slice);
-            } catch {
-              break;
-            }
-          }
-        }
-      }
-    }
-    return void 0;
-  }
-}
-function resolveGatewayModel(modelId) {
-  const models = getModels(GATEWAY_PROVIDER);
-  const matched = models.find((model) => model.id === modelId);
-  if (!matched) {
-    throw new Error(`Unknown AI Gateway model id: ${modelId}`);
-  }
-  return matched;
-}
-async function completeText(params) {
-  const model = resolveGatewayModel(params.modelId);
-  const apiKey = getPiGatewayApiKeyOverride();
-  const requestMessagesAttribute = serializeGenAiAttribute(params.messages);
-  const systemInstructionsAttribute = params.system ? serializeGenAiAttribute([{ type: "text", content: params.system }]) : void 0;
-  const startAttributes = {
-    "gen_ai.provider.name": GEN_AI_PROVIDER_NAME,
-    "gen_ai.operation.name": GEN_AI_OPERATION_CHAT,
-    "gen_ai.request.model": params.modelId,
-    ...systemInstructionsAttribute ? { "gen_ai.system_instructions": systemInstructionsAttribute } : {},
-    ...requestMessagesAttribute ? { "gen_ai.input.messages": requestMessagesAttribute } : {},
-    "app.ai.auth_mode": apiKey ? "oidc" : "api_key"
-  };
-  setSpanAttributes(startAttributes);
-  const message = await completeSimple(
-    model,
-    {
-      systemPrompt: params.system,
-      messages: params.messages
-    },
-    {
-      ...apiKey ? { apiKey } : {},
-      temperature: params.temperature,
-      maxTokens: params.maxTokens,
-      signal: params.signal,
-      metadata: params.metadata
-    }
-  );
-  const outputText = extractText(message);
-  const outputMessagesAttribute = serializeGenAiAttribute([
-    {
-      role: "assistant",
-      content: outputText ? [{ type: "text", text: outputText }] : []
-    }
-  ]);
-  const usageAttributes = extractGenAiUsageAttributes(message);
-  const endAttributes = {
-    "gen_ai.provider.name": GEN_AI_PROVIDER_NAME,
-    "gen_ai.operation.name": GEN_AI_OPERATION_CHAT,
-    "gen_ai.request.model": params.modelId,
-    ...outputMessagesAttribute ? { "gen_ai.output.messages": outputMessagesAttribute } : {},
-    ...usageAttributes,
-    ...message.stopReason ? { "gen_ai.response.finish_reasons": [message.stopReason] } : {}
-  };
-  setSpanAttributes(endAttributes);
-  if (message.stopReason === "error") {
-    const providerMessage = message.errorMessage?.trim() || "Unknown provider error";
-    logWarn(
-      "ai_completion_provider_error",
-      {},
-      {
-        "gen_ai.provider.name": GEN_AI_PROVIDER_NAME,
-        "gen_ai.operation.name": GEN_AI_OPERATION_CHAT,
-        "gen_ai.request.model": params.modelId,
-        "error.message": providerMessage
-      },
-      "AI completion returned provider error"
-    );
-    throw new Error(`AI provider error: ${providerMessage}`);
-  }
-  return {
-    message,
-    text: outputText
-  };
-}
-async function completeObject(params) {
-  const startedAt = Date.now();
-  let text = "";
-  try {
-    ({ text } = await completeText({
-      modelId: params.modelId,
-      system: params.system,
-      temperature: params.temperature,
-      maxTokens: params.maxTokens,
-      signal: params.signal,
-      metadata: params.metadata,
-      messages: [
-        {
-          role: "user",
-          content: params.prompt,
-          timestamp: Date.now()
-        }
-      ]
-    }));
-  } catch (error) {
-    logException(
-      error,
-      "ai_completion_failed",
-      {},
-      {
-        "gen_ai.provider.name": GEN_AI_PROVIDER_NAME,
-        "gen_ai.operation.name": GEN_AI_OPERATION_CHAT,
-        "gen_ai.request.model": params.modelId,
-        "app.ai.duration_ms": Date.now() - startedAt
-      },
-      "AI object completion failed"
-    );
-    throw error;
-  }
-  const candidate = parseJsonCandidate(text);
-  const parsed = params.schema.safeParse(candidate);
-  if (!parsed.success) {
-    const preview = text.length > 400 ? `${text.slice(0, 400)}...` : text;
-    logWarn(
-      "ai_completion_schema_parse_failed",
-      {},
-      {
-        "gen_ai.provider.name": GEN_AI_PROVIDER_NAME,
-        "gen_ai.operation.name": GEN_AI_OPERATION_CHAT,
-        "gen_ai.request.model": params.modelId,
-        "app.ai.duration_ms": Date.now() - startedAt,
-        "app.ai.response_preview": preview
-      },
-      "AI object completion schema parse failed"
-    );
-    throw new Error(
-      `Model did not return valid JSON for schema: ${parsed.error.message}. Raw response: ${preview}`
-    );
-  }
-  return {
-    object: parsed.data,
-    text
-  };
-}
-
 // src/chat/slack/message.ts
 function getSlackMessageTs(message) {
   if (message.id.endsWith(":message_changed_mention") && message.raw && typeof message.raw === "object") {
@@ -3354,7 +3137,6 @@ function buildSystemPrompt(params) {
       [
         "- For factual or external questions, run tools/skills first, then answer from evidence.",
         "- Use tool descriptions as the source of truth for when each tool should or should not be called.",
-        "- Use `reportProgress` only for sparse, meaningful progress updates. Pass a short user-facing status message, and do not call it for every tool or small substep.",
         "- When using CLI tools through `bash`, prefer deterministic non-interactive flags and avoid commands that wait for prompts or editors.",
         "- Keep routine setup and research steps silent in user-facing replies. Do not narrate duplicate checks, credential issuance, file writes, or similar internal progress unless the result is user-relevant.",
         "- If a routine prerequisite check finds nothing notable, omit it entirely from the final reply and report only the user-relevant outcome.",
@@ -5068,11 +4850,11 @@ function createReadFileTool() {
 import { Type as Type6 } from "@sinclair/typebox";
 function createReportProgressTool() {
   return tool({
-    description: "Update assistant status with a short user-facing progress message. Use this sparingly for meaningful progress changes, not for every tool call or minor substep.",
+    description: "Update the user-visible assistant loading message with a short progress phase. For every non-trivial turn, call this early with the initial major work phase, then call it again only when the major phase meaningfully changes. Messages must be written in sentence case with a present-participle verb (e.g. 'Searching docs', 'Reviewing results', 'Running checks'). Skip trivial direct answers, generic filler, and minor substeps.",
     inputSchema: Type6.Object({
       message: Type6.String({
         minLength: 1,
-        description: "Short user-facing progress message. The UI truncates it if needed."
+        description: "Short user-facing progress message."
       })
     })
   });
@@ -5346,7 +5128,7 @@ function createOperationKey(toolName, input) {
 // src/chat/tools/slack/channel-post-message.ts
 function createSlackChannelPostMessageTool(context, state) {
   return tool({
-    description: "Post a message in the active Slack channel context (outside the thread). Use this when the user explicitly asks to post/send/share/say something in the channel. Do not use for normal thread replies or speculative broadcasts. Do not claim a channel message was posted unless this tool succeeds in this turn.",
+    description: "Post a message in the active Slack channel context (outside the thread). Use this only when the user explicitly asks to post/send/share/say something in the current channel. Do not use it for normal thread replies, speculative broadcasts, or requests targeting another named channel; explain that limitation instead. Do not claim a channel message was posted unless this tool succeeds in this turn.",
     inputSchema: Type9.Object({
       text: Type9.String({
         minLength: 1,
@@ -8715,7 +8497,7 @@ function handleToolExecutionError(error, toolName, toolCallId, shouldTrace, trac
 }
 
 // src/chat/tools/agent-tools.ts
-function createAgentTools(tools, sandbox, spanContext, onStatus, sandboxExecutor, capabilityRuntime, pluginAuthOrchestration, hooks) {
+function createAgentTools(tools, sandbox, spanContext, onStatus, sandboxExecutor, capabilityRuntime, pluginAuthOrchestration, onToolCall) {
   const shouldTrace = shouldEmitDevAgentTrace();
   return Object.entries(tools).map(([toolName, toolDef]) => ({
     name: toolName,
@@ -8725,7 +8507,7 @@ function createAgentTools(tools, sandbox, spanContext, onStatus, sandboxExecutor
     execute: async (toolCallId, params) => {
       const normalizedToolCallId = typeof toolCallId === "string" && toolCallId.length > 0 ? toolCallId : void 0;
       const toolArgumentsAttribute = serializeGenAiAttribute(params);
-      hooks?.onToolCall?.(toolName);
+      onToolCall?.(toolName);
       const traceToolContext = {
         ...spanContext,
         conversationId: spanContext.conversationId,
@@ -8843,7 +8625,7 @@ function isExecutionEscapeResponse(text) {
   if (!trimmed) return false;
   return isExecutionDeferralResponse(trimmed) || isToolAccessDisclaimerResponse(trimmed);
 }
-function parseJsonCandidate2(text) {
+function parseJsonCandidate(text) {
   const trimmed = text.trim();
   if (!trimmed) return void 0;
   try {
@@ -8871,7 +8653,7 @@ function isToolPayloadShape(payload) {
   return false;
 }
 function isRawToolPayloadResponse(text) {
-  const parsed = parseJsonCandidate2(text);
+  const parsed = parseJsonCandidate(text);
   if (Array.isArray(parsed)) {
     return parsed.some((entry) => isToolPayloadShape(entry));
   }
@@ -9139,6 +8921,7 @@ function buildTurnResult(input) {
     shouldTrace,
     spanContext,
     usage,
+    thinkingSelection,
     correlation,
     assistantUserName
   } = input;
@@ -9214,6 +8997,7 @@ function buildTurnResult(input) {
     outcome: resolvedOutcome,
     modelId: botConfig.modelId,
     assistantMessageCount: assistantMessages.length,
+    thinkingLevel: thinkingSelection.thinkingLevel,
     toolCalls,
     toolResultCount: toolResults.length,
     toolErrorCount,
@@ -9236,6 +9020,120 @@ function buildTurnResult(input) {
   };
 }
 
+// src/chat/services/turn-thinking-level.ts
+import { z } from "zod";
+var CLASSIFIER_CONFIDENCE_THRESHOLD = 0.75;
+var MAX_ROUTER_CONTEXT_CHARS = 1200;
+var TURN_THINKING_LEVELS = ["none", "low", "medium", "high"];
+var turnExecutionProfileSchema = z.object({
+  thinking_level: z.enum(TURN_THINKING_LEVELS),
+  confidence: z.number().min(0).max(1),
+  reason: z.string().min(1)
+});
+var DEFAULT_THINKING_LEVEL = "low";
+function trimContextForRouter(text) {
+  const trimmed = text?.trim();
+  if (!trimmed) {
+    return void 0;
+  }
+  return trimmed.length <= MAX_ROUTER_CONTEXT_CHARS ? trimmed : trimmed.slice(-MAX_ROUTER_CONTEXT_CHARS);
+}
+function buildClassifierSystemPrompt() {
+  return [
+    "You route assistant turns to the cheapest thinking level that is still likely to succeed.",
+    "Choose exactly one bucket: none, low, medium, or high.",
+    "",
+    "Use none for greetings, acknowledgments, and trivial single-step asks.",
+    "Use low for straightforward explanations or simple one-step work.",
+    "Use medium for investigations, ambiguous asks, multi-step analysis, or likely multi-tool work.",
+    "Use high for code changes, debugging/root-cause analysis, research-heavy work, non-trivial drafting, or explicit requests to be thorough.",
+    "",
+    "Return JSON only with thinking_level, confidence, and reason."
+  ].join("\n");
+}
+function buildClassifierPrompt(args) {
+  const sections = [];
+  const context = trimContextForRouter(args.conversationContext);
+  if (context) {
+    sections.push("<thread-background>", context, "</thread-background>", "");
+  }
+  sections.push(
+    "<turn-context>",
+    `- active_skills: ${args.activeSkillNames.join(", ") || "none"}`,
+    `- attachment_count: ${args.attachmentCount}`,
+    "</turn-context>",
+    "",
+    '<current-instruction priority="highest">',
+    args.messageText.trim() || "[empty]",
+    "</current-instruction>"
+  );
+  for (const block of args.currentTurnBlocks ?? []) {
+    const trimmed = block.trim();
+    if (!trimmed) {
+      continue;
+    }
+    sections.push("", trimmed);
+  }
+  return sections.join("\n");
+}
+async function selectTurnThinkingLevel(args) {
+  const activeSkillNames = [...new Set(args.activeSkillNames ?? [])].sort();
+  try {
+    const result = await args.completeObject({
+      modelId: args.fastModelId,
+      schema: turnExecutionProfileSchema,
+      maxTokens: 120,
+      metadata: {
+        modelId: args.fastModelId,
+        threadId: args.context?.threadId ?? "",
+        channelId: args.context?.channelId ?? "",
+        requesterId: args.context?.requesterId ?? "",
+        runId: args.context?.runId ?? ""
+      },
+      prompt: buildClassifierPrompt({
+        activeSkillNames,
+        attachmentCount: args.attachmentCount ?? 0,
+        conversationContext: args.conversationContext,
+        currentTurnBlocks: args.currentTurnBlocks,
+        messageText: args.messageText
+      }),
+      thinkingLevel: "low",
+      system: buildClassifierSystemPrompt(),
+      temperature: 0
+    });
+    const parsed = turnExecutionProfileSchema.parse(result.object);
+    if (parsed.confidence < CLASSIFIER_CONFIDENCE_THRESHOLD) {
+      return {
+        confidence: parsed.confidence,
+        thinkingLevel: DEFAULT_THINKING_LEVEL,
+        reason: `low_confidence_default:${parsed.reason.trim()}`
+      };
+    }
+    return {
+      confidence: parsed.confidence,
+      thinkingLevel: parsed.thinking_level,
+      reason: parsed.reason.trim()
+    };
+  } catch {
+    return {
+      thinkingLevel: DEFAULT_THINKING_LEVEL,
+      reason: "classifier_error_default"
+    };
+  }
+}
+function toAgentThinkingLevel(level) {
+  switch (level) {
+    case "none":
+      return "off";
+    case "low":
+      return "low";
+    case "medium":
+      return "medium";
+    case "high":
+      return "high";
+  }
+}
+
 // src/chat/state/turn-session-store.ts
 var AGENT_TURN_SESSION_PREFIX = "junior:agent_turn_session";
 var AGENT_TURN_SESSION_TTL_MS = 24 * 60 * 60 * 1e3;
@@ -9513,6 +9411,7 @@ function createMcpAuthOrchestration(deps, abortAgent) {
 
 // src/chat/respond.ts
 var startupDiscoveryLogged = false;
+var MAX_ROUTER_ATTACHMENT_PREVIEW_CHARS = 2e3;
 function buildOmittedImageAttachmentNotice(count) {
   return [
     "<omitted-image-attachments>",
@@ -9523,6 +9422,89 @@ function buildOmittedImageAttachmentNotice(count) {
     "</omitted-image-attachments>"
   ].join("\n");
 }
+function trimRouterAttachmentText(text) {
+  const normalized = text.replaceAll("\0", " ").trim();
+  if (!normalized) {
+    return "";
+  }
+  return normalized.length <= MAX_ROUTER_ATTACHMENT_PREVIEW_CHARS ? normalized : `${normalized.slice(0, MAX_ROUTER_ATTACHMENT_PREVIEW_CHARS)}...`;
+}
+function supportsRouterTextPreview(mediaType) {
+  const baseMediaType = mediaType.split(";", 1)[0]?.trim().toLowerCase();
+  if (!baseMediaType) {
+    return false;
+  }
+  return baseMediaType.startsWith("text/") || baseMediaType === "application/json" || baseMediaType === "application/xml" || baseMediaType === "application/x-www-form-urlencoded" || baseMediaType.endsWith("+json") || baseMediaType.endsWith("+xml");
+}
+function buildRouterAttachmentBlock(attachment) {
+  if (attachment.promptText) {
+    return trimRouterAttachmentText(attachment.promptText);
+  }
+  const header = [
+    "<attachment>",
+    `filename: ${attachment.filename ?? "unnamed"}`,
+    `media_type: ${attachment.mediaType}`
+  ];
+  if (attachment.data && supportsRouterTextPreview(attachment.mediaType)) {
+    const preview = trimRouterAttachmentText(attachment.data.toString("utf8"));
+    if (preview) {
+      return [
+        ...header,
+        "<text-preview>",
+        preview,
+        "</text-preview>",
+        "</attachment>"
+      ].join("\n");
+    }
+  }
+  return [...header, "</attachment>"].join("\n");
+}
+function buildUserTurnInput(args) {
+  const routerBlocks = [];
+  const userContentParts = [
+    { type: "text", text: args.userTurnText }
+  ];
+  if (args.omittedImageAttachmentCount > 0) {
+    const omittedImagesNotice = buildOmittedImageAttachmentNotice(
+      args.omittedImageAttachmentCount
+    );
+    userContentParts.push({ type: "text", text: omittedImagesNotice });
+    routerBlocks.push(omittedImagesNotice);
+  }
+  for (const attachment of args.userAttachments ?? []) {
+    routerBlocks.push(buildRouterAttachmentBlock(attachment));
+    if (attachment.promptText) {
+      userContentParts.push({
+        type: "text",
+        text: attachment.promptText
+      });
+      continue;
+    }
+    if (attachment.mediaType.startsWith("image/")) {
+      if (!attachment.data) {
+        throw new Error("Image attachment is missing image data");
+      }
+      userContentParts.push({
+        type: "image",
+        data: attachment.data.toString("base64"),
+        mimeType: attachment.mediaType
+      });
+      continue;
+    }
+    if (!attachment.data) {
+      throw new Error("Attachment is missing attachment data");
+    }
+    userContentParts.push({
+      type: "text",
+      text: encodeNonImageAttachmentForPrompt({
+        data: attachment.data,
+        mediaType: attachment.mediaType,
+        filename: attachment.filename
+      })
+    });
+  }
+  return { routerBlocks, userContentParts };
+}
 function mcpToolsToDefinitions(mcpTools) {
   const defs = {};
   for (const tool2 of mcpTools) {
@@ -9550,6 +9532,7 @@ async function generateAssistantReply(messageText, context = {}) {
   let sandboxExecutor;
   let timedOut = false;
   let turnUsage;
+  let thinkingSelection;
   const getSandboxMetadata = () => sandboxExecutor ? {
     sandboxId: sandboxExecutor.getSandboxId(),
     sandboxDependencyProfileHash: sandboxExecutor.getDependencyProfileHash()
@@ -9728,6 +9711,34 @@ async function generateAssistantReply(messageText, context = {}) {
         turnContext: { traceId: getActiveTraceId() }
       }
     );
+    const { routerBlocks, userContentParts } = buildUserTurnInput({
+      omittedImageAttachmentCount: context.omittedImageAttachmentCount ?? 0,
+      userAttachments: context.userAttachments,
+      userTurnText
+    });
+    thinkingSelection = await selectTurnThinkingLevel({
+      activeSkillNames: activeSkills.map((skill) => skill.name),
+      attachmentCount: context.userAttachments?.length,
+      completeObject,
+      conversationContext: context.conversationContext,
+      context: {
+        threadId: context.correlation?.threadId,
+        channelId: context.correlation?.channelId,
+        requesterId: context.correlation?.requesterId,
+        runId: context.correlation?.runId
+      },
+      currentTurnBlocks: routerBlocks,
+      fastModelId: botConfig.fastModelId,
+      messageText: userInput
+    });
+    setSpanAttributes({
+      "gen_ai.request.model": botConfig.modelId,
+      "app.ai.reasoning_effort": thinkingSelection.thinkingLevel,
+      "app.ai.thinking_level_reason": thinkingSelection.reason,
+      ...thinkingSelection.confidence !== void 0 ? {
+        "app.ai.thinking_level_confidence": thinkingSelection.confidence
+      } : {}
+    });
     timeoutResumeMessages = [];
     const generatedFiles = [];
     const replyFiles = [];
@@ -9889,44 +9900,6 @@ async function generateAssistantReply(messageText, context = {}) {
       runtimeMetadata: getRuntimeMetadata(),
       threadParticipants: context.threadParticipants
     });
-    const userContentParts = [{ type: "text", text: userTurnText }];
-    const omittedImageAttachmentCount = context.omittedImageAttachmentCount ?? 0;
-    if (omittedImageAttachmentCount > 0) {
-      userContentParts.push({
-        type: "text",
-        text: buildOmittedImageAttachmentNotice(omittedImageAttachmentCount)
-      });
-    }
-    for (const attachment of context.userAttachments ?? []) {
-      if (attachment.promptText) {
-        userContentParts.push({
-          type: "text",
-          text: attachment.promptText
-        });
-      } else if (attachment.mediaType.startsWith("image/")) {
-        if (!attachment.data) {
-          throw new Error("Image attachment is missing image data");
-        }
-        userContentParts.push({
-          type: "image",
-          data: attachment.data.toString("base64"),
-          mimeType: attachment.mediaType
-        });
-      } else {
-        if (!attachment.data) {
-          throw new Error("Attachment is missing attachment data");
-        }
-        const promptAttachment = {
-          data: attachment.data,
-          mediaType: attachment.mediaType,
-          filename: attachment.filename
-        };
-        userContentParts.push({
-          type: "text",
-          text: encodeNonImageAttachmentForPrompt(promptAttachment)
-        });
-      }
-    }
     const inputMessagesAttribute = serializeGenAiAttribute([
       {
         role: "system",
@@ -9937,21 +9910,8 @@ async function generateAssistantReply(messageText, context = {}) {
         content: userContentParts.map((part) => toObservablePromptPart(part))
       }
     ]);
-    const agentToolHooks = {
-      onToolCall: (toolName) => {
-        toolCalls.push(toolName);
-        Promise.resolve(context.onToolCall?.(toolName)).catch((error) => {
-          logWarn(
-            "streaming_tool_call_error",
-            {},
-            {
-              "error.message": error instanceof Error ? error.message : String(error),
-              "gen_ai.tool.name": toolName
-            },
-            "Failed to deliver tool call event to stream coordinator"
-          );
-        });
-      }
+    const onToolCall = (toolName) => {
+      toolCalls.push(toolName);
     };
     const baseAgentTools = createAgentTools(
       tools,
@@ -9961,7 +9921,7 @@ async function generateAssistantReply(messageText, context = {}) {
       sandboxExecutor,
       capabilityRuntime,
       pluginAuth,
-      agentToolHooks
+      onToolCall
     );
     const agentTools = [...baseAgentTools];
     const syncMcpAgentTools = () => {
@@ -9975,7 +9935,7 @@ async function generateAssistantReply(messageText, context = {}) {
         sandboxExecutor,
         capabilityRuntime,
         pluginAuth,
-        agentToolHooks
+        onToolCall
       );
       agentTools.length = 0;
       agentTools.push(...baseAgentTools, ...mcpAgentTools);
@@ -9986,6 +9946,7 @@ async function generateAssistantReply(messageText, context = {}) {
       initialState: {
         systemPrompt: baseInstructions,
         model: resolveGatewayModel(botConfig.modelId),
+        thinkingLevel: toAgentThinkingLevel(thinkingSelection.thinkingLevel),
         tools: agentTools
       }
     });
@@ -10031,7 +9992,7 @@ async function generateAssistantReply(messageText, context = {}) {
     let completedAssistantTurn = false;
     try {
       if (resumedFromCheckpoint) {
-        agent.replaceMessages(existingCheckpoint.piMessages);
+        agent.state.messages = existingCheckpoint.piMessages;
       }
       beforeMessageCount = agent.state.messages.length;
       await withSpan(
@@ -10073,6 +10034,9 @@ async function generateAssistantReply(messageText, context = {}) {
                   "gen_ai.provider.name": GEN_AI_PROVIDER_NAME,
                   "gen_ai.operation.name": "invoke_agent",
                   "gen_ai.request.model": botConfig.modelId,
+                  ...thinkingSelection ? {
+                    "app.ai.reasoning_effort": thinkingSelection.thinkingLevel
+                  } : {},
                   "app.ai.turn_timeout_ms": botConfig.turnTimeoutMs
                 },
                 "Agent turn timed out and was aborted"
@@ -10117,6 +10081,7 @@ async function generateAssistantReply(messageText, context = {}) {
           "gen_ai.provider.name": GEN_AI_PROVIDER_NAME,
           "gen_ai.operation.name": "invoke_agent",
           "gen_ai.request.model": botConfig.modelId,
+          "app.ai.reasoning_effort": thinkingSelection.thinkingLevel,
           ...inputMessagesAttribute ? { "gen_ai.input.messages": inputMessagesAttribute } : {}
         }
       );
@@ -10148,6 +10113,7 @@ async function generateAssistantReply(messageText, context = {}) {
       shouldTrace,
       spanContext,
       usage: turnUsage,
+      thinkingSelection,
       correlation: context.correlation,
       assistantUserName: context.assistant?.userName
     });
@@ -10234,6 +10200,9 @@ async function generateAssistantReply(messageText, context = {}) {
         outcome: "provider_error",
         modelId: botConfig.modelId,
         assistantMessageCount: 0,
+        ...thinkingSelection ? {
+          thinkingLevel: thinkingSelection.thinkingLevel
+        } : {},
         toolCalls: [],
         toolResultCount: 0,
         toolErrorCount: 0,
@@ -10672,11 +10641,10 @@ function buildSlackReplyFooter(args) {
       value: formatSlackDuration(durationMs)
     });
   }
-  const traceId = args.traceId?.trim();
-  if (traceId) {
+  if (args.thinkingLevel) {
     items.push({
-      label: "Trace",
-      value: traceId
+      label: "Thinking",
+      value: args.thinkingLevel
     });
   }
   return items.length > 0 ? { items } : void 0;
@@ -10999,7 +10967,7 @@ async function resumeSlackTurn(args) {
     const footer = buildSlackReplyFooter({
       conversationId: args.replyContext?.correlation?.conversationId ?? lockKey,
       durationMs: reply.diagnostics.durationMs,
-      traceId: getActiveTraceId(),
+      thinkingLevel: reply.diagnostics.thinkingLevel,
       usage: reply.diagnostics.usage
     });
     await postSlackApiReplyPosts({
@@ -12296,14 +12264,14 @@ async function POST(request, waitUntil) {
 }
 
 // src/chat/services/subscribed-decision.ts
-import { z } from "zod";
-var replyDecisionSchema = z.object({
-  should_reply: z.boolean().describe("Whether Junior should respond to this thread message."),
-  should_unsubscribe: z.boolean().optional().describe(
+import { z as z2 } from "zod";
+var replyDecisionSchema = z2.object({
+  should_reply: z2.boolean().describe("Whether Junior should respond to this thread message."),
+  should_unsubscribe: z2.boolean().optional().describe(
     "Whether Junior should unsubscribe from this thread because the user clearly asked it to stop participating."
   ),
-  confidence: z.number().min(0).max(1).describe("Classifier confidence from 0 to 1."),
-  reason: z.string().optional().describe("Short reason for the decision.")
+  confidence: z2.number().min(0).max(1).describe("Classifier confidence from 0 to 1."),
+  reason: z2.string().optional().describe("Short reason for the decision.")
 });
 var ROUTER_CONFIDENCE_THRESHOLD = 0.8;
 var LEADING_SLACK_MENTION_RE = /^\s*<@([A-Z0-9]+)(?:\|([^>]+))?>[\s,:-]*/i;
@@ -13993,7 +13961,7 @@ function createReplyToThread(deps) {
             slackChannelId: channelId,
             runId,
             assistantUserName: botConfig.userName,
-            modelId: botConfig.modelId
+            modelId: reply.diagnostics.modelId
           };
           const diagnosticsAttributes = {
             "gen_ai.provider.name": GEN_AI_PROVIDER_NAME,
@@ -14004,6 +13972,9 @@ function createReplyToThread(deps) {
             "app.ai.tool_error_results": reply.diagnostics.toolErrorCount,
             "app.ai.tool_call_count": reply.diagnostics.toolCalls.length,
             "app.ai.used_primary_text": reply.diagnostics.usedPrimaryText,
+            ...reply.diagnostics.thinkingLevel ? {
+              "app.ai.reasoning_effort": reply.diagnostics.thinkingLevel
+            } : {},
             ...reply.diagnostics.stopReason ? {
               "gen_ai.response.finish_reasons": [
                 reply.diagnostics.stopReason
@@ -14065,7 +14036,7 @@ function createReplyToThread(deps) {
           const replyFooter = buildSlackReplyFooter({
             conversationId,
             durationMs: reply.diagnostics.durationMs,
-            traceId: getActiveTraceId(),
+            thinkingLevel: reply.diagnostics.thinkingLevel,
             usage: reply.diagnostics.usage
           });
           const shouldUseSlackFooter = Boolean(replyFooter) && Boolean(channelId && threadTs) && thread.adapter?.name === "slack";

package/dist/{chunk-4PVJHUEV.js → chunk-LEYD42MR.js}

@@ -1,6 +1,11 @@
 import {
+  extractGenAiUsageAttributes,
   getPluginRuntimeDependencies,
   getPluginRuntimePostinstall,
+  logException,
+  logWarn,
+  serializeGenAiAttribute,
+  setSpanAttributes,
   withSpan
 } from "./chunk-RZJDO55D.js";
 
@@ -8,6 +13,9 @@ import {
 import { createMemoryState } from "@chat-adapter/state-memory";
 import { createRedisState } from "@chat-adapter/state-redis";
 
+// src/chat/config.ts
+import { getModel } from "@mariozechner/pi-ai";
+
 // src/chat/optional-string.ts
 function toOptionalTrimmed(value) {
   if (!value) {
@@ -17,6 +25,233 @@ function toOptionalTrimmed(value) {
   return trimmed.length > 0 ? trimmed : void 0;
 }
 
+// src/chat/pi/client.ts
+import {
+  completeSimple,
+  getEnvApiKey,
+  getModels,
+  registerApiProvider
+} from "@mariozechner/pi-ai";
+import {
+  streamAnthropic,
+  streamSimpleAnthropic
+} from "@mariozechner/pi-ai/anthropic";
+registerApiProvider({
+  api: "anthropic-messages",
+  stream: streamAnthropic,
+  streamSimple: streamSimpleAnthropic
+});
+var GATEWAY_PROVIDER = "vercel-ai-gateway";
+var GEN_AI_PROVIDER_NAME = GATEWAY_PROVIDER;
+var GEN_AI_OPERATION_CHAT = "chat";
+var MISSING_GATEWAY_CREDENTIALS_ERROR = "Missing AI gateway credentials (AI_GATEWAY_API_KEY or VERCEL_OIDC_TOKEN)";
+function getGatewayApiKey() {
+  return toOptionalTrimmed(getEnvApiKey("vercel-ai-gateway")) ?? toOptionalTrimmed(process.env.VERCEL_OIDC_TOKEN);
+}
+function getPiGatewayApiKeyOverride() {
+  return toOptionalTrimmed(process.env.VERCEL_OIDC_TOKEN);
+}
+function extractText(message) {
+  return (message.content ?? []).filter((part) => part.type === "text" && typeof part.text === "string").map((part) => part.text ?? "").join("").trim();
+}
+function parseJsonCandidate(text) {
+  const trimmed = text.trim();
+  if (!trimmed) return void 0;
+  try {
+    return JSON.parse(trimmed);
+  } catch {
+    const fencedBlocks = [
+      ...trimmed.matchAll(/```(?:json)?\s*([\s\S]*?)\s*```/gi)
+    ];
+    for (const block of fencedBlocks) {
+      try {
+        return JSON.parse(block[1]);
+      } catch {
+      }
+    }
+    const openBraceIndex = trimmed.indexOf("{");
+    if (openBraceIndex >= 0) {
+      let depth = 0;
+      let inString = false;
+      let escaped = false;
+      for (let index = openBraceIndex; index < trimmed.length; index += 1) {
+        const char = trimmed[index];
+        if (inString) {
+          if (escaped) {
+            escaped = false;
+            continue;
+          }
+          if (char === "\\") {
+            escaped = true;
+            continue;
+          }
+          if (char === '"') {
+            inString = false;
+          }
+          continue;
+        }
+        if (char === '"') {
+          inString = true;
+          continue;
+        }
+        if (char === "{") {
+          depth += 1;
+          continue;
+        }
+        if (char === "}") {
+          depth -= 1;
+          if (depth === 0) {
+            const slice = trimmed.slice(openBraceIndex, index + 1);
+            try {
+              return JSON.parse(slice);
+            } catch {
+              break;
+            }
+          }
+        }
+      }
+    }
+    return void 0;
+  }
+}
+function resolveGatewayModel(modelId) {
+  const matched = getModels(GATEWAY_PROVIDER).find(
+    (model) => model.id === modelId
+  );
+  if (!matched) {
+    throw new Error(`Unknown AI Gateway model id: ${modelId}`);
+  }
+  return matched;
+}
+async function completeText(params) {
+  const model = resolveGatewayModel(params.modelId);
+  const apiKey = getPiGatewayApiKeyOverride();
+  const requestMessagesAttribute = serializeGenAiAttribute(params.messages);
+  const systemInstructionsAttribute = params.system ? serializeGenAiAttribute([{ type: "text", content: params.system }]) : void 0;
+  const startAttributes = {
+    "gen_ai.provider.name": GEN_AI_PROVIDER_NAME,
+    "gen_ai.operation.name": GEN_AI_OPERATION_CHAT,
+    "gen_ai.request.model": params.modelId,
+    ...systemInstructionsAttribute ? { "gen_ai.system_instructions": systemInstructionsAttribute } : {},
+    ...requestMessagesAttribute ? { "gen_ai.input.messages": requestMessagesAttribute } : {},
+    "app.ai.auth_mode": apiKey ? "oidc" : "api_key",
+    ...params.thinkingLevel ? { "app.ai.reasoning_effort": params.thinkingLevel } : {}
+  };
+  setSpanAttributes(startAttributes);
+  const message = await completeSimple(
+    model,
+    {
+      systemPrompt: params.system,
+      messages: params.messages
+    },
+    {
+      ...apiKey ? { apiKey } : {},
+      temperature: params.temperature,
+      maxTokens: params.maxTokens,
+      reasoning: params.thinkingLevel,
+      signal: params.signal,
+      metadata: params.metadata
+    }
+  );
+  const outputText = extractText(message);
+  const outputMessagesAttribute = serializeGenAiAttribute([
+    {
+      role: "assistant",
+      content: outputText ? [{ type: "text", text: outputText }] : []
+    }
+  ]);
+  const usageAttributes = extractGenAiUsageAttributes(message);
+  const endAttributes = {
+    "gen_ai.provider.name": GEN_AI_PROVIDER_NAME,
+    "gen_ai.operation.name": GEN_AI_OPERATION_CHAT,
+    "gen_ai.request.model": params.modelId,
+    ...outputMessagesAttribute ? { "gen_ai.output.messages": outputMessagesAttribute } : {},
+    ...usageAttributes,
+    ...message.stopReason ? { "gen_ai.response.finish_reasons": [message.stopReason] } : {},
+    ...params.thinkingLevel ? { "app.ai.reasoning_effort": params.thinkingLevel } : {}
+  };
+  setSpanAttributes(endAttributes);
+  if (message.stopReason === "error") {
+    const providerMessage = message.errorMessage?.trim() || "Unknown provider error";
+    logWarn(
+      "ai_completion_provider_error",
+      {},
+      {
+        "gen_ai.provider.name": GEN_AI_PROVIDER_NAME,
+        "gen_ai.operation.name": GEN_AI_OPERATION_CHAT,
+        "gen_ai.request.model": params.modelId,
+        "error.message": providerMessage
+      },
+      "AI completion returned provider error"
+    );
+    throw new Error(`AI provider error: ${providerMessage}`);
+  }
+  return {
+    message,
+    text: outputText
+  };
+}
+async function completeObject(params) {
+  const startedAt = Date.now();
+  let text = "";
+  try {
+    ({ text } = await completeText({
+      modelId: params.modelId,
+      system: params.system,
+      thinkingLevel: params.thinkingLevel,
+      temperature: params.temperature,
+      maxTokens: params.maxTokens,
+      signal: params.signal,
+      metadata: params.metadata,
+      messages: [
+        {
+          role: "user",
+          content: params.prompt,
+          timestamp: Date.now()
+        }
+      ]
+    }));
+  } catch (error) {
+    logException(
+      error,
+      "ai_completion_failed",
+      {},
+      {
+        "gen_ai.provider.name": GEN_AI_PROVIDER_NAME,
+        "gen_ai.operation.name": GEN_AI_OPERATION_CHAT,
+        "gen_ai.request.model": params.modelId,
+        "app.ai.duration_ms": Date.now() - startedAt
+      },
+      "AI object completion failed"
+    );
+    throw error;
+  }
+  const candidate = parseJsonCandidate(text);
+  const parsed = params.schema.safeParse(candidate);
+  if (!parsed.success) {
+    const preview = text.length > 400 ? `${text.slice(0, 400)}...` : text;
+    logWarn(
+      "ai_completion_schema_parse_failed",
+      {},
+      {
+        "gen_ai.provider.name": GEN_AI_PROVIDER_NAME,
+        "gen_ai.operation.name": GEN_AI_OPERATION_CHAT,
+        "gen_ai.request.model": params.modelId,
+        "app.ai.duration_ms": Date.now() - startedAt,
+        "app.ai.response_preview": preview
+      },
+      "AI object completion schema parse failed"
+    );
+    throw new Error(
+      `Model did not return valid JSON for schema: ${parsed.error.message}. Raw response: ${preview}`
+    );
+  }
+  return {
+    object: parsed.data,
+    text
+  };
+}
+
 // src/chat/config.ts
 var MIN_AGENT_TURN_TIMEOUT_MS = 10 * 1e3;
 var DEFAULT_AGENT_TURN_TIMEOUT_MS = 12 * 60 * 1e3;
@@ -79,15 +314,26 @@ function parseLoadingMessages(rawValue) {
     return value.trim();
   });
 }
+var DEFAULT_MODEL_ID = getModel("vercel-ai-gateway", "openai/gpt-5.4").id;
+var DEFAULT_FAST_MODEL_ID = getModel(
+  "vercel-ai-gateway",
+  "openai/gpt-5.4-mini"
+).id;
+function validateGatewayModelId(raw) {
+  const trimmed = toOptionalTrimmed(raw);
+  if (trimmed === void 0) return void 0;
+  resolveGatewayModel(trimmed);
+  return trimmed;
+}
 function readBotConfig(env) {
   const functionMaxDurationSeconds = resolveFunctionMaxDurationSeconds(env);
   const maxTurnTimeoutMs = resolveMaxTurnTimeoutMs(functionMaxDurationSeconds);
   return {
     userName: env.JUNIOR_BOT_NAME ?? "junior",
-    modelId: env.AI_MODEL ?? "anthropic/claude-sonnet-4.6",
-    fastModelId: env.AI_FAST_MODEL ?? env.AI_MODEL ?? "anthropic/claude-haiku-4.5",
+    modelId: validateGatewayModelId(env.AI_MODEL) ?? DEFAULT_MODEL_ID,
+    fastModelId: validateGatewayModelId(env.AI_FAST_MODEL ?? env.AI_MODEL) ?? DEFAULT_FAST_MODEL_ID,
     loadingMessages: parseLoadingMessages(env.JUNIOR_LOADING_MESSAGES),
-    visionModelId: toOptionalTrimmed(env.AI_VISION_MODEL),
+    visionModelId: validateGatewayModelId(env.AI_VISION_MODEL),
     turnTimeoutMs: parseAgentTurnTimeoutMs(
       env.AGENT_TURN_TIMEOUT_MS,
       maxTurnTimeoutMs
@@ -842,7 +1088,13 @@ function isSnapshotMissingError(error) {
 }
 
 export {
-  toOptionalTrimmed,
+  GEN_AI_PROVIDER_NAME,
+  MISSING_GATEWAY_CREDENTIALS_ERROR,
+  getGatewayApiKey,
+  getPiGatewayApiKeyOverride,
+  resolveGatewayModel,
+  completeText,
+  completeObject,
   botConfig,
   getSlackBotToken,
   getSlackSigningSecret,

package/dist/cli/snapshot-warmup.js

@@ -1,7 +1,7 @@
 import {
   disconnectStateAdapter,
   resolveRuntimeDependencySnapshot
-} from "../chunk-4PVJHUEV.js";
+} from "../chunk-LEYD42MR.js";
 import {
   getPluginProviders,
   getPluginRuntimeDependencies,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@sentry/junior",
-  "version": "0.27.2",
+  "version": "0.29.0",
   "private": false,
   "publishConfig": {
     "access": "public"
@@ -25,8 +25,8 @@
     "@chat-adapter/state-memory": "4.26.0",
     "@chat-adapter/state-redis": "4.26.0",
     "@logtape/logtape": "^2.0.5",
-    "@mariozechner/pi-agent-core": "0.59.0",
-    "@mariozechner/pi-ai": "0.59.0",
+    "@mariozechner/pi-agent-core": "0.68.1",
+    "@mariozechner/pi-ai": "0.68.1",
     "@modelcontextprotocol/sdk": "1.29.0",
     "@sinclair/typebox": "^0.34.49",
     "@slack/web-api": "^7.15.1",