@agentv/core 3.11.0 → 3.12.0

package/dist/index.js

@@ -19,7 +19,7 @@ import {
   readTextFile,
   resolveFileReference,
   resolveTargetDefinition
-} from "./chunk-AVTN5AB7.js";
+} from "./chunk-4XWPXNQM.js";
 import {
   AgentvProvider
 } from "./chunk-W5YDZWT4.js";
@@ -28,7 +28,7 @@ import {
 } from "./chunk-HFSYZHGF.js";
 import {
   SimpleTraceFileExporter
-} from "./chunk-HMXZ2AX4.js";
+} from "./chunk-3G2KXH7N.js";
 
 // src/evaluation/trace.ts
 function computeTraceSummary(messages) {
@@ -6793,265 +6793,7 @@ var MockProvider = class {
   }
 };
 
-// src/evaluation/providers/pi-utils.ts
-function extractPiTextContent(content) {
-  if (typeof content === "string") {
-    return content;
-  }
-  if (!Array.isArray(content)) {
-    return void 0;
-  }
-  const textParts = [];
-  for (const part of content) {
-    if (!part || typeof part !== "object") {
-      continue;
-    }
-    const p = part;
-    if (p.type === "text" && typeof p.text === "string") {
-      textParts.push(p.text);
-    }
-  }
-  return textParts.length > 0 ? textParts.join("\n") : void 0;
-}
-function toFiniteNumber(value) {
-  if (typeof value === "number" && Number.isFinite(value)) return value;
-  return void 0;
-}
-
-// src/evaluation/providers/pi-agent-sdk.ts
-var piAgentModule = null;
-var piAiModule = null;
-async function loadPiModules() {
-  if (!piAgentModule || !piAiModule) {
-    try {
-      [piAgentModule, piAiModule] = await Promise.all([
-        import("@mariozechner/pi-agent-core"),
-        import("@mariozechner/pi-ai")
-      ]);
-    } catch (error) {
-      throw new Error(
-        `Failed to load pi-agent-sdk dependencies. Please install them:
-  npm install @mariozechner/pi-agent-core @mariozechner/pi-ai
-
-Original error: ${error instanceof Error ? error.message : String(error)}`
-      );
-    }
-  }
-  return {
-    Agent: piAgentModule.Agent,
-    getModel: piAiModule.getModel,
-    getEnvApiKey: piAiModule.getEnvApiKey
-  };
-}
-var PiAgentSdkProvider = class {
-  id;
-  kind = "pi-agent-sdk";
-  targetName;
-  supportsBatch = false;
-  config;
-  constructor(targetName, config) {
-    this.id = `pi-agent-sdk:${targetName}`;
-    this.targetName = targetName;
-    this.config = config;
-  }
-  async invoke(request) {
-    if (request.signal?.aborted) {
-      throw new Error("Pi agent SDK request was aborted before execution");
-    }
-    const { Agent, getModel, getEnvApiKey } = await loadPiModules();
-    const startTimeIso = (/* @__PURE__ */ new Date()).toISOString();
-    const startMs = Date.now();
-    const providerName = this.config.subprovider ?? "anthropic";
-    const modelId = this.config.model ?? "claude-sonnet-4-20250514";
-    const model = getModel(providerName, modelId);
-    const systemPrompt = this.config.systemPrompt ?? "Answer directly and concisely.";
-    const agent = new Agent({
-      initialState: {
-        systemPrompt,
-        model,
-        tools: [],
-        // No tools for simple Q&A
-        messages: []
-      },
-      getApiKey: async (provider) => {
-        return this.config.apiKey ?? getEnvApiKey(provider) ?? void 0;
-      }
-    });
-    let tokenUsage;
-    let costUsd;
-    const toolTrackers = /* @__PURE__ */ new Map();
-    const completedToolResults = /* @__PURE__ */ new Map();
-    const unsubscribe = agent.subscribe((event) => {
-      switch (event.type) {
-        case "message_end": {
-          const msg = event.message;
-          if (msg && typeof msg === "object" && "role" in msg && msg.role === "assistant" && "usage" in msg) {
-            const usage = msg.usage;
-            if (usage && typeof usage === "object") {
-              const u = usage;
-              const input = toFiniteNumber(u.input);
-              const output = toFiniteNumber(u.output);
-              const cached = toFiniteNumber(u.cacheRead);
-              let callDelta;
-              if (input !== void 0 || output !== void 0) {
-                callDelta = {
-                  input: input ?? 0,
-                  output: output ?? 0,
-                  ...cached !== void 0 ? { cached } : {}
-                };
-                tokenUsage = {
-                  input: (tokenUsage?.input ?? 0) + callDelta.input,
-                  output: (tokenUsage?.output ?? 0) + callDelta.output,
-                  ...cached !== void 0 ? { cached: (tokenUsage?.cached ?? 0) + cached } : tokenUsage?.cached !== void 0 ? { cached: tokenUsage.cached } : {}
-                };
-              }
-              const cost = u.cost;
-              if (cost && typeof cost === "object") {
-                const total = toFiniteNumber(cost.total);
-                if (total !== void 0) {
-                  costUsd = (costUsd ?? 0) + total;
-                }
-              }
-              request.streamCallbacks?.onLlmCallEnd?.(modelId, callDelta);
-            }
-          }
-          break;
-        }
-        case "tool_execution_start": {
-          toolTrackers.set(event.toolCallId, {
-            toolCallId: event.toolCallId,
-            toolName: event.toolName,
-            args: event.args,
-            startMs: Date.now(),
-            startTime: (/* @__PURE__ */ new Date()).toISOString()
-          });
-          request.streamCallbacks?.onToolCallStart?.(event.toolName, event.toolCallId);
-          break;
-        }
-        case "tool_execution_end": {
-          const tracker = toolTrackers.get(event.toolCallId);
-          const durationMs = tracker ? Date.now() - tracker.startMs : 0;
-          completedToolResults.set(event.toolCallId, {
-            output: event.result,
-            durationMs
-          });
-          request.streamCallbacks?.onToolCallEnd?.(
-            event.toolName,
-            tracker?.args,
-            event.result,
-            durationMs,
-            event.toolCallId
-          );
-          toolTrackers.delete(event.toolCallId);
-          break;
-        }
-      }
-    });
-    try {
-      if (this.config.timeoutMs) {
-        const timeoutMs = this.config.timeoutMs;
-        const timeoutPromise = new Promise((_, reject) => {
-          setTimeout(
-            () => reject(new Error(`Pi agent SDK timed out after ${timeoutMs}ms`)),
-            timeoutMs
-          );
-        });
-        await Promise.race([agent.prompt(request.question), timeoutPromise]);
-      } else {
-        await agent.prompt(request.question);
-      }
-      await agent.waitForIdle();
-      const agentMessages = agent.state.messages;
-      const output = [];
-      for (const msg of agentMessages) {
-        output.push(convertAgentMessage(msg, toolTrackers, completedToolResults));
-      }
-      const endTimeIso = (/* @__PURE__ */ new Date()).toISOString();
-      const durationMs = Date.now() - startMs;
-      return {
-        raw: {
-          messages: agentMessages,
-          systemPrompt,
-          model: this.config.model,
-          subprovider: this.config.subprovider
-        },
-        output,
-        tokenUsage,
-        costUsd,
-        durationMs,
-        startTime: startTimeIso,
-        endTime: endTimeIso
-      };
-    } finally {
-      unsubscribe();
-    }
-  }
-};
-function convertAgentMessage(message, toolTrackers, completedToolResults) {
-  if (!message || typeof message !== "object") {
-    return { role: "unknown", content: String(message) };
-  }
-  const msg = message;
-  const role = typeof msg.role === "string" ? msg.role : "unknown";
-  const content = extractPiTextContent(msg.content);
-  const toolCalls = extractToolCalls3(msg.content, toolTrackers, completedToolResults);
-  const startTime = typeof msg.timestamp === "number" ? new Date(msg.timestamp).toISOString() : typeof msg.timestamp === "string" ? msg.timestamp : void 0;
-  let msgTokenUsage;
-  if (msg.usage && typeof msg.usage === "object") {
-    const u = msg.usage;
-    const input = toFiniteNumber(u.input);
-    const output = toFiniteNumber(u.output);
-    if (input !== void 0 || output !== void 0) {
-      msgTokenUsage = {
-        input: input ?? 0,
-        output: output ?? 0,
-        ...toFiniteNumber(u.cacheRead) !== void 0 ? { cached: toFiniteNumber(u.cacheRead) } : {}
-      };
-    }
-  }
-  const metadata = {};
-  if (msg.api) metadata.api = msg.api;
-  if (msg.provider) metadata.provider = msg.provider;
-  if (msg.model) metadata.model = msg.model;
-  if (msg.stopReason) metadata.stopReason = msg.stopReason;
-  return {
-    role,
-    content,
-    toolCalls: toolCalls.length > 0 ? toolCalls : void 0,
-    startTime,
-    metadata: Object.keys(metadata).length > 0 ? metadata : void 0,
-    tokenUsage: msgTokenUsage
-  };
-}
-function extractToolCalls3(content, toolTrackers, completedToolResults) {
-  if (!Array.isArray(content)) {
-    return [];
-  }
-  const toolCalls = [];
-  for (const part of content) {
-    if (!part || typeof part !== "object") {
-      continue;
-    }
-    const p = part;
-    if (p.type === "toolCall" && typeof p.name === "string") {
-      const id = typeof p.id === "string" ? p.id : void 0;
-      const tracker = id ? toolTrackers.get(id) : void 0;
-      const completed = id ? completedToolResults.get(id) : void 0;
-      toolCalls.push({
-        tool: p.name,
-        input: p.arguments,
-        id,
-        output: completed?.output,
-        durationMs: completed?.durationMs,
-        startTime: tracker?.startTime,
-        endTime: tracker?.startTime && completed?.durationMs !== void 0 ? new Date(new Date(tracker.startTime).getTime() + completed.durationMs).toISOString() : void 0
-      });
-    }
-  }
-  return toolCalls;
-}
-
-// src/evaluation/providers/pi-coding-agent.ts
+// src/evaluation/providers/pi-cli.ts
 import { spawn as spawn3 } from "node:child_process";
 import { randomUUID as randomUUID7 } from "node:crypto";
 import { createWriteStream as createWriteStream5 } from "node:fs";
@@ -7112,25 +6854,50 @@ function subscribeToPiLogEntries(listener) {
   };
 }
 
-// src/evaluation/providers/pi-coding-agent.ts
+// src/evaluation/providers/pi-utils.ts
+function extractPiTextContent(content) {
+  if (typeof content === "string") {
+    return content;
+  }
+  if (!Array.isArray(content)) {
+    return void 0;
+  }
+  const textParts = [];
+  for (const part of content) {
+    if (!part || typeof part !== "object") {
+      continue;
+    }
+    const p = part;
+    if (p.type === "text" && typeof p.text === "string") {
+      textParts.push(p.text);
+    }
+  }
+  return textParts.length > 0 ? textParts.join("\n") : void 0;
+}
+function toFiniteNumber(value) {
+  if (typeof value === "number" && Number.isFinite(value)) return value;
+  return void 0;
+}
+
+// src/evaluation/providers/pi-cli.ts
 var WORKSPACE_PREFIX = "agentv-pi-";
 var PROMPT_FILENAME = "prompt.md";
-var PiCodingAgentProvider = class {
+var PiCliProvider = class {
   id;
-  kind = "pi-coding-agent";
+  kind = "pi-cli";
   targetName;
   supportsBatch = false;
   config;
   runPi;
   constructor(targetName, config, runner = defaultPiRunner) {
-    this.id = `pi-coding-agent:${targetName}`;
+    this.id = `pi-cli:${targetName}`;
     this.targetName = targetName;
     this.config = config;
     this.runPi = runner;
   }
   async invoke(request) {
     if (request.signal?.aborted) {
-      throw new Error("Pi coding agent request was aborted before execution");
+      throw new Error("Pi CLI request was aborted before execution");
     }
     const inputFiles = normalizeInputFiles(request.inputFiles);
     const startTime = (/* @__PURE__ */ new Date()).toISOString();
@@ -7140,17 +6907,17 @@ var PiCodingAgentProvider = class {
     try {
       const promptFile = path17.join(workspaceRoot, PROMPT_FILENAME);
       await writeFile(promptFile, request.question, "utf8");
-      const args = this.buildPiArgs(request.question, inputFiles, request.captureFileChanges);
+      const args = this.buildPiArgs(request.question, inputFiles);
       const cwd = this.resolveCwd(workspaceRoot, request.cwd);
       const result = await this.executePi(args, cwd, request.signal, logger);
       if (result.timedOut) {
         throw new Error(
-          `Pi coding agent timed out${formatTimeoutSuffix3(this.config.timeoutMs ?? void 0)}`
+          `Pi CLI timed out${formatTimeoutSuffix3(this.config.timeoutMs ?? void 0)}`
         );
       }
       if (result.exitCode !== 0) {
         const detail = pickDetail(result.stderr, result.stdout);
-        const prefix = `Pi coding agent exited with code ${result.exitCode}`;
+        const prefix = `Pi CLI exited with code ${result.exitCode}`;
         throw new Error(detail ? `${prefix}: ${detail}` : prefix);
       }
       const parsed = parsePiJsonl(result.stdout);
@@ -7207,7 +6974,7 @@ var PiCodingAgentProvider = class {
     }
     return path17.resolve(this.config.cwd);
   }
-  buildPiArgs(prompt, inputFiles, _captureFileChanges) {
+  buildPiArgs(prompt, inputFiles) {
     const args = [];
     if (this.config.subprovider) {
       args.push("--provider", this.config.subprovider);
@@ -7259,7 +7026,7 @@ ${prompt}` : prompt;
       const err = error;
       if (err.code === "ENOENT") {
         throw new Error(
-          `Pi coding agent executable '${this.config.executable}' was not found. Update the target settings.executable or add it to PATH.`
+          `Pi CLI executable '${this.config.executable}' was not found. Update the target executable or add it to PATH.`
         );
       }
       throw error;
@@ -7269,26 +7036,18 @@ ${prompt}` : prompt;
     const env = { ...process.env };
     if (this.config.apiKey) {
       const provider = this.config.subprovider?.toLowerCase() ?? "google";
-      switch (provider) {
-        case "google":
-        case "gemini":
-          env.GEMINI_API_KEY = this.config.apiKey;
-          break;
-        case "anthropic":
-          env.ANTHROPIC_API_KEY = this.config.apiKey;
-          break;
-        case "openai":
-          env.OPENAI_API_KEY = this.config.apiKey;
-          break;
-        case "groq":
-          env.GROQ_API_KEY = this.config.apiKey;
-          break;
-        case "xai":
-          env.XAI_API_KEY = this.config.apiKey;
-          break;
-        case "openrouter":
-          env.OPENROUTER_API_KEY = this.config.apiKey;
-          break;
+      const ENV_KEY_MAP = {
+        google: "GEMINI_API_KEY",
+        gemini: "GEMINI_API_KEY",
+        anthropic: "ANTHROPIC_API_KEY",
+        openai: "OPENAI_API_KEY",
+        groq: "GROQ_API_KEY",
+        xai: "XAI_API_KEY",
+        openrouter: "OPENROUTER_API_KEY"
+      };
+      const envKey = ENV_KEY_MAP[provider];
+      if (envKey) {
+        env[envKey] = this.config.apiKey;
       }
     }
     return env;
@@ -7306,7 +7065,7 @@ ${prompt}` : prompt;
     if (this.config.logDir) {
       return path17.resolve(this.config.logDir);
     }
-    return path17.join(process.cwd(), ".agentv", "logs", "pi-coding-agent");
+    return path17.join(process.cwd(), ".agentv", "logs", "pi-cli");
   }
   async createStreamLogger(request) {
     const logDir = this.resolveLogDirectory();
@@ -7358,7 +7117,7 @@ var PiStreamLogger = class _PiStreamLogger {
   static async create(options) {
     const logger = new _PiStreamLogger(options.filePath, options.format);
     const header = [
-      "# Pi Coding Agent stream log",
+      "# Pi CLI stream log",
       `# target: ${options.targetName}`,
       options.evalCaseId ? `# eval: ${options.evalCaseId}` : void 0,
       options.attempt !== void 0 ? `# attempt: ${options.attempt + 1}` : void 0,
@@ -7507,10 +7266,10 @@ function summarizePiEvent(event) {
       return `${type}: ${role}`;
     }
     case "message_update": {
-      const event2 = record.assistantMessageEvent;
-      const eventType = event2?.type;
+      const evt = record.assistantMessageEvent;
+      const eventType = evt?.type;
       if (eventType === "text_delta") {
-        const delta = event2?.delta;
+        const delta = evt?.delta;
         if (typeof delta === "string") {
           const preview = delta.length > 50 ? `${delta.slice(0, 50)}...` : delta;
           return `text_delta: ${preview}`;
@@ -7532,7 +7291,7 @@ function tryParseJsonValue(rawLine) {
 function parsePiJsonl(output) {
   const trimmed = output.trim();
   if (trimmed.length === 0) {
-    throw new Error("Pi coding agent produced no output");
+    throw new Error("Pi CLI produced no output");
   }
   const lines = trimmed.split(/\r?\n/).map((line) => line.trim()).filter((line) => line.length > 0);
   const parsed = [];
@@ -7543,38 +7302,27 @@ function parsePiJsonl(output) {
     }
   }
   if (parsed.length === 0) {
-    throw new Error("Pi coding agent produced no valid JSON output");
+    throw new Error("Pi CLI produced no valid JSON output");
   }
   return parsed;
 }
 function extractMessages(events) {
   for (let i = events.length - 1; i >= 0; i--) {
     const event = events[i];
-    if (!event || typeof event !== "object") {
-      continue;
-    }
+    if (!event || typeof event !== "object") continue;
     const record = event;
-    if (record.type !== "agent_end") {
-      continue;
-    }
+    if (record.type !== "agent_end") continue;
     const messages = record.messages;
-    if (!Array.isArray(messages)) {
-      continue;
-    }
+    if (!Array.isArray(messages)) continue;
     return messages.map(convertPiMessage).filter((m) => m !== void 0);
   }
   const output = [];
   for (const event of events) {
-    if (!event || typeof event !== "object") {
-      continue;
-    }
+    if (!event || typeof event !== "object") continue;
     const record = event;
     if (record.type === "turn_end") {
-      const message = record.message;
-      const converted = convertPiMessage(message);
-      if (converted) {
-        output.push(converted);
-      }
+      const converted = convertPiMessage(record.message);
+      if (converted) output.push(converted);
     }
   }
   return output;
@@ -7591,10 +7339,7 @@ function extractTokenUsage(events) {
       const input = toFiniteNumber(u.input_tokens ?? u.inputTokens ?? u.input);
       const output = toFiniteNumber(u.output_tokens ?? u.outputTokens ?? u.output);
       if (input !== void 0 || output !== void 0) {
-        const result = {
-          input: input ?? 0,
-          output: output ?? 0
-        };
+        const result = { input: input ?? 0, output: output ?? 0 };
         const cached = toFiniteNumber(u.cache_read_input_tokens ?? u.cached ?? u.cachedTokens);
         const reasoning = toFiniteNumber(u.reasoning_tokens ?? u.reasoningTokens ?? u.reasoning);
         return {
@@ -7634,40 +7379,577 @@ function aggregateUsageFromMessages(messages) {
       }
     }
   }
-  if (!found) return void 0;
-  const result = { input: totalInput, output: totalOutput };
-  if (totalCached !== void 0) {
-    return { ...result, cached: totalCached };
+  if (!found) return void 0;
+  const result = { input: totalInput, output: totalOutput };
+  if (totalCached !== void 0) {
+    return { ...result, cached: totalCached };
+  }
+  return result;
+}
+function convertPiMessage(message) {
+  if (!message || typeof message !== "object") return void 0;
+  const msg = message;
+  const role = msg.role;
+  if (typeof role !== "string") return void 0;
+  const content = extractPiTextContent(msg.content);
+  const toolCalls = extractToolCalls3(msg.content);
+  const startTime = typeof msg.timestamp === "number" ? new Date(msg.timestamp).toISOString() : typeof msg.timestamp === "string" ? msg.timestamp : void 0;
+  const metadata = {};
+  if (msg.api) metadata.api = msg.api;
+  if (msg.provider) metadata.provider = msg.provider;
+  if (msg.model) metadata.model = msg.model;
+  if (msg.usage) metadata.usage = msg.usage;
+  if (msg.stopReason) metadata.stopReason = msg.stopReason;
+  return {
+    role,
+    content,
+    toolCalls: toolCalls.length > 0 ? toolCalls : void 0,
+    startTime,
+    metadata: Object.keys(metadata).length > 0 ? metadata : void 0
+  };
+}
+function extractToolCalls3(content) {
+  if (!Array.isArray(content)) return [];
+  const toolCalls = [];
+  for (const part of content) {
+    if (!part || typeof part !== "object") continue;
+    const p = part;
+    if (p.type === "tool_use" && typeof p.name === "string") {
+      toolCalls.push({
+        tool: p.name,
+        input: p.input,
+        id: typeof p.id === "string" ? p.id : void 0
+      });
+    }
+    if (p.type === "toolCall" && typeof p.name === "string") {
+      toolCalls.push({
+        tool: p.name,
+        input: p.arguments,
+        id: typeof p.id === "string" ? p.id : void 0
+      });
+    }
+    if (p.type === "tool_result" && typeof p.tool_use_id === "string") {
+      const existing = toolCalls.find((tc) => tc.id === p.tool_use_id);
+      if (existing) {
+        const idx = toolCalls.indexOf(existing);
+        toolCalls[idx] = { ...existing, output: p.content };
+      }
+    }
+  }
+  return toolCalls;
+}
+function escapeAtSymbols(prompt) {
+  return prompt.replace(/@\[([^\]]+)\]:/g, "[[$1]]:");
+}
+function pickDetail(stderr, stdout) {
+  const errorText = stderr.trim();
+  if (errorText.length > 0) return errorText;
+  const stdoutText = stdout.trim();
+  return stdoutText.length > 0 ? stdoutText : void 0;
+}
+function formatTimeoutSuffix3(timeoutMs) {
+  if (!timeoutMs || timeoutMs <= 0) return "";
+  return ` after ${Math.ceil(timeoutMs / 1e3)}s`;
+}
+async function defaultPiRunner(options) {
+  return await new Promise((resolve, reject) => {
+    const parts = options.executable.split(/\s+/);
+    const executable = parts[0];
+    const executableArgs = parts.slice(1);
+    const allArgs = [...executableArgs, ...options.args];
+    const child = spawn3(executable, allArgs, {
+      cwd: options.cwd,
+      env: options.env,
+      stdio: ["pipe", "pipe", "pipe"],
+      shell: false
+    });
+    let stdout = "";
+    let stderr = "";
+    let timedOut = false;
+    const onAbort = () => {
+      child.kill("SIGTERM");
+    };
+    if (options.signal) {
+      if (options.signal.aborted) {
+        onAbort();
+      } else {
+        options.signal.addEventListener("abort", onAbort, { once: true });
+      }
+    }
+    let timeoutHandle;
+    if (options.timeoutMs && options.timeoutMs > 0) {
+      timeoutHandle = setTimeout(() => {
+        timedOut = true;
+        child.kill("SIGTERM");
+      }, options.timeoutMs);
+      timeoutHandle.unref?.();
+    }
+    child.stdout.setEncoding("utf8");
+    child.stdout.on("data", (chunk) => {
+      stdout += chunk;
+      options.onStdoutChunk?.(chunk);
+    });
+    child.stderr.setEncoding("utf8");
+    child.stderr.on("data", (chunk) => {
+      stderr += chunk;
+      options.onStderrChunk?.(chunk);
+    });
+    child.stdin.end();
+    const cleanup = () => {
+      if (timeoutHandle) clearTimeout(timeoutHandle);
+      if (options.signal) options.signal.removeEventListener("abort", onAbort);
+    };
+    child.on("error", (error) => {
+      cleanup();
+      reject(error);
+    });
+    child.on("close", (code) => {
+      cleanup();
+      resolve({
+        stdout,
+        stderr,
+        exitCode: typeof code === "number" ? code : -1,
+        timedOut
+      });
+    });
+  });
+}
+
+// src/evaluation/providers/pi-coding-agent.ts
+import { execSync } from "node:child_process";
+import { randomUUID as randomUUID8 } from "node:crypto";
+import { createWriteStream as createWriteStream6 } from "node:fs";
+import { mkdir as mkdir7 } from "node:fs/promises";
+import path18 from "node:path";
+import { createInterface } from "node:readline";
+var piCodingAgentModule = null;
+var piAiModule = null;
+async function promptInstall() {
+  if (!process.stdout.isTTY) return false;
+  const rl = createInterface({ input: process.stdin, output: process.stderr });
+  try {
+    return await new Promise((resolve) => {
+      rl.question(
+        "@mariozechner/pi-coding-agent is not installed. Install it now? (y/N) ",
+        (answer) => resolve(answer.trim().toLowerCase() === "y")
+      );
+    });
+  } finally {
+    rl.close();
+  }
+}
+async function loadSdkModules() {
+  if (!piCodingAgentModule || !piAiModule) {
+    try {
+      [piCodingAgentModule, piAiModule] = await Promise.all([
+        import("@mariozechner/pi-coding-agent"),
+        import("@mariozechner/pi-ai")
+      ]);
+    } catch {
+      if (await promptInstall()) {
+        console.error("Installing @mariozechner/pi-coding-agent...");
+        execSync("bun add @mariozechner/pi-coding-agent", { stdio: "inherit" });
+        [piCodingAgentModule, piAiModule] = await Promise.all([
+          import("@mariozechner/pi-coding-agent"),
+          import("@mariozechner/pi-ai")
+        ]);
+      } else {
+        throw new Error(
+          "pi-coding-agent SDK is not installed. Install it with:\n  bun add @mariozechner/pi-coding-agent"
+        );
+      }
+    }
+  }
+  const toolMap = {
+    read: piCodingAgentModule.readTool,
+    bash: piCodingAgentModule.bashTool,
+    edit: piCodingAgentModule.editTool,
+    write: piCodingAgentModule.writeTool,
+    grep: piCodingAgentModule.grepTool,
+    find: piCodingAgentModule.findTool,
+    ls: piCodingAgentModule.lsTool
+  };
+  return {
+    createAgentSession: piCodingAgentModule.createAgentSession,
+    codingTools: piCodingAgentModule.codingTools,
+    toolMap,
+    SessionManager: piCodingAgentModule.SessionManager,
+    getModel: piAiModule.getModel
+  };
+}
+var PiCodingAgentProvider = class {
+  id;
+  kind = "pi-coding-agent";
+  targetName;
+  supportsBatch = false;
+  config;
+  constructor(targetName, config) {
+    this.id = `pi-coding-agent:${targetName}`;
+    this.targetName = targetName;
+    this.config = config;
+  }
+  async invoke(request) {
+    if (request.signal?.aborted) {
+      throw new Error("Pi coding agent request was aborted before execution");
+    }
+    const inputFiles = normalizeInputFiles(request.inputFiles);
+    const startTime = (/* @__PURE__ */ new Date()).toISOString();
+    const startMs = Date.now();
+    const sdk = await loadSdkModules();
+    const logger = await this.createStreamLogger(request).catch(() => void 0);
+    try {
+      const cwd = this.resolveCwd(request.cwd);
+      const providerName = this.config.subprovider ?? "google";
+      const modelId = this.config.model ?? "gemini-2.5-flash";
+      this.setApiKeyEnv(providerName);
+      const model = sdk.getModel(providerName, modelId);
+      const tools = this.resolveTools(sdk);
+      const { session } = await sdk.createAgentSession({
+        cwd,
+        model,
+        tools,
+        thinkingLevel: this.config.thinking,
+        sessionManager: sdk.SessionManager.inMemory(cwd)
+      });
+      let tokenUsage;
+      let costUsd;
+      const toolTrackers = /* @__PURE__ */ new Map();
+      const completedToolResults = /* @__PURE__ */ new Map();
+      const unsubscribe = session.subscribe((event) => {
+        logger?.handleEvent(event);
+        switch (event.type) {
+          case "message_end": {
+            const msg = event.message;
+            if (msg && typeof msg === "object" && "role" in msg && msg.role === "assistant" && "usage" in msg) {
+              const usage = msg.usage;
+              if (usage && typeof usage === "object") {
+                const u = usage;
+                const input = toFiniteNumber(u.input);
+                const output = toFiniteNumber(u.output);
+                const cached = toFiniteNumber(u.cacheRead);
+                let callDelta;
+                if (input !== void 0 || output !== void 0) {
+                  callDelta = {
+                    input: input ?? 0,
+                    output: output ?? 0,
+                    ...cached !== void 0 ? { cached } : {}
+                  };
+                  tokenUsage = {
+                    input: (tokenUsage?.input ?? 0) + callDelta.input,
+                    output: (tokenUsage?.output ?? 0) + callDelta.output,
+                    ...cached !== void 0 ? { cached: (tokenUsage?.cached ?? 0) + cached } : tokenUsage?.cached !== void 0 ? { cached: tokenUsage.cached } : {}
+                  };
+                }
+                const cost = u.cost;
+                if (cost && typeof cost === "object") {
+                  const total = toFiniteNumber(cost.total);
+                  if (total !== void 0) {
+                    costUsd = (costUsd ?? 0) + total;
+                  }
+                }
+                request.streamCallbacks?.onLlmCallEnd?.(modelId, callDelta);
+              }
+            }
+            break;
+          }
+          case "tool_execution_start": {
+            toolTrackers.set(event.toolCallId, {
+              toolCallId: event.toolCallId,
+              toolName: event.toolName,
+              args: event.args,
+              startMs: Date.now(),
+              startTime: (/* @__PURE__ */ new Date()).toISOString()
+            });
+            request.streamCallbacks?.onToolCallStart?.(event.toolName, event.toolCallId);
+            break;
+          }
+          case "tool_execution_end": {
+            const tracker = toolTrackers.get(event.toolCallId);
+            const durationMs = tracker ? Date.now() - tracker.startMs : 0;
+            completedToolResults.set(event.toolCallId, {
+              output: event.result,
+              durationMs
+            });
+            request.streamCallbacks?.onToolCallEnd?.(
+              event.toolName,
+              tracker?.args,
+              event.result,
+              durationMs,
+              event.toolCallId
+            );
+            toolTrackers.delete(event.toolCallId);
+            break;
+          }
+        }
+      });
+      try {
+        const systemPrompt = this.config.systemPrompt;
+        let prompt = request.question;
+        if (systemPrompt) {
+          prompt = `${systemPrompt}
+
+${prompt}`;
+        }
+        if (inputFiles && inputFiles.length > 0) {
+          const fileList = inputFiles.map((f) => `@${f}`).join("\n");
+          prompt = `${prompt}
+
+Files:
+${fileList}`;
+        }
+        if (this.config.timeoutMs) {
+          const timeoutMs = this.config.timeoutMs;
+          let timeoutId;
+          const timeoutPromise = new Promise((_, reject) => {
+            timeoutId = setTimeout(
+              () => reject(
+                new Error(`Pi coding agent timed out after ${Math.ceil(timeoutMs / 1e3)}s`)
+              ),
+              timeoutMs
+            );
+          });
+          try {
+            await Promise.race([session.prompt(prompt), timeoutPromise]);
+          } finally {
+            if (timeoutId !== void 0) clearTimeout(timeoutId);
+          }
+        } else {
+          await session.prompt(prompt);
+        }
+        const agentMessages = session.agent.state.messages;
+        const output = [];
+        for (const msg of agentMessages) {
+          output.push(convertAgentMessage(msg, toolTrackers, completedToolResults));
+        }
+        const endTime = (/* @__PURE__ */ new Date()).toISOString();
+        const durationMs = Date.now() - startMs;
+        return {
+          raw: {
+            messages: agentMessages,
+            model: this.config.model,
+            provider: this.config.subprovider
+          },
+          output,
+          tokenUsage,
+          costUsd,
+          durationMs,
+          startTime,
+          endTime
+        };
+      } finally {
+        unsubscribe();
+        session.dispose();
+      }
+    } finally {
+      await logger?.close();
+    }
+  }
+  /** Maps config apiKey to the provider-specific env var the SDK reads. */
+  setApiKeyEnv(providerName) {
+    if (!this.config.apiKey) return;
+    const ENV_KEY_MAP = {
+      google: "GEMINI_API_KEY",
+      gemini: "GEMINI_API_KEY",
+      anthropic: "ANTHROPIC_API_KEY",
+      openai: "OPENAI_API_KEY",
+      groq: "GROQ_API_KEY",
+      xai: "XAI_API_KEY",
+      openrouter: "OPENROUTER_API_KEY"
+    };
+    const envKey = ENV_KEY_MAP[providerName.toLowerCase()];
+    if (envKey) {
+      process.env[envKey] = this.config.apiKey;
+    }
+  }
+  resolveCwd(cwdOverride) {
+    if (cwdOverride) {
+      return path18.resolve(cwdOverride);
+    }
+    if (this.config.cwd) {
+      return path18.resolve(this.config.cwd);
+    }
+    return process.cwd();
+  }
+  resolveTools(sdk) {
+    if (!this.config.tools) {
+      return sdk.codingTools;
+    }
+    const toolNames = this.config.tools.split(",").map((t) => t.trim().toLowerCase());
+    const selected = [];
+    for (const name of toolNames) {
+      if (name in sdk.toolMap) {
+        selected.push(sdk.toolMap[name]);
+      }
+    }
+    return selected.length > 0 ? selected : sdk.codingTools;
+  }
+  resolveLogDirectory() {
+    if (this.config.logDir) {
+      return path18.resolve(this.config.logDir);
+    }
+    return path18.join(process.cwd(), ".agentv", "logs", "pi-coding-agent");
+  }
+  async createStreamLogger(request) {
+    const logDir = this.resolveLogDirectory();
+    if (!logDir) {
+      return void 0;
+    }
+    try {
+      await mkdir7(logDir, { recursive: true });
+    } catch (error) {
+      const message = error instanceof Error ? error.message : String(error);
+      console.warn(`Skipping Pi stream logging (could not create ${logDir}): ${message}`);
+      return void 0;
+    }
+    const filePath = path18.join(logDir, buildLogFilename6(request, this.targetName));
+    try {
+      const logger = await PiStreamLogger2.create({
+        filePath,
+        targetName: this.targetName,
+        evalCaseId: request.evalCaseId,
+        attempt: request.attempt,
+        format: this.config.logFormat ?? "summary"
+      });
+      recordPiLogEntry({
+        filePath,
+        targetName: this.targetName,
+        evalCaseId: request.evalCaseId,
+        attempt: request.attempt
+      });
+      return logger;
+    } catch (error) {
+      const message = error instanceof Error ? error.message : String(error);
+      console.warn(`Skipping Pi stream logging for ${filePath}: ${message}`);
+      return void 0;
+    }
+  }
+};
+var PiStreamLogger2 = class _PiStreamLogger {
+  filePath;
+  stream;
+  startedAt = Date.now();
+  format;
+  constructor(filePath, format) {
+    this.filePath = filePath;
+    this.format = format;
+    this.stream = createWriteStream6(filePath, { flags: "a" });
+  }
+  static async create(options) {
+    const logger = new _PiStreamLogger(options.filePath, options.format);
+    const header = [
+      "# Pi Coding Agent stream log",
+      `# target: ${options.targetName}`,
+      options.evalCaseId ? `# eval: ${options.evalCaseId}` : void 0,
+      options.attempt !== void 0 ? `# attempt: ${options.attempt + 1}` : void 0,
+      `# started: ${(/* @__PURE__ */ new Date()).toISOString()}`,
+      ""
+    ].filter((line) => Boolean(line));
+    for (const line of header) {
+      logger.stream.write(`${line}
+`);
+    }
+    return logger;
+  }
+  handleEvent(event) {
+    if (!event || typeof event !== "object") return;
+    const record = event;
+    const type = typeof record.type === "string" ? record.type : void 0;
+    if (!type) return;
+    const message = this.format === "json" ? JSON.stringify(event, null, 2) : summarizeSdkEvent2(event);
+    if (message) {
+      this.stream.write(`[+${formatElapsed6(this.startedAt)}] ${message}
+`);
+    }
+  }
+  async close() {
+    await new Promise((resolve, reject) => {
+      this.stream.once("error", reject);
+      this.stream.end(() => resolve());
+    });
+  }
+};
+function summarizeSdkEvent2(event) {
+  if (!event || typeof event !== "object") return void 0;
+  const record = event;
+  const type = typeof record.type === "string" ? record.type : void 0;
+  if (!type) return void 0;
+  switch (type) {
+    case "agent_start":
+    case "agent_end":
+    case "turn_start":
+    case "turn_end":
+      return type;
+    case "message_start":
+    case "message_end": {
+      const msg = record.message;
+      return `${type}: ${msg?.role ?? "unknown"}`;
+    }
+    case "tool_execution_start":
+      return `tool_start: ${record.toolName}`;
+    case "tool_execution_end":
+      return `tool_end: ${record.toolName}`;
+    default:
+      return type;
+  }
+}
+function buildLogFilename6(request, targetName) {
+  const timestamp = (/* @__PURE__ */ new Date()).toISOString().replace(/[:.]/g, "-");
+  const evalId = sanitizeForFilename6(request.evalCaseId ?? "pi");
+  const attemptSuffix = request.attempt !== void 0 ? `_attempt-${request.attempt + 1}` : "";
+  const target = sanitizeForFilename6(targetName);
+  return `${timestamp}_${target}_${evalId}${attemptSuffix}_${randomUUID8().slice(0, 8)}.log`;
+}
+function sanitizeForFilename6(value) {
+  const sanitized = value.replace(/[^A-Za-z0-9._-]+/g, "_");
+  return sanitized.length > 0 ? sanitized : "pi";
+}
+function formatElapsed6(startedAt) {
+  const elapsedSeconds = Math.floor((Date.now() - startedAt) / 1e3);
+  const hours = Math.floor(elapsedSeconds / 3600);
+  const minutes = Math.floor(elapsedSeconds % 3600 / 60);
+  const seconds = elapsedSeconds % 60;
+  if (hours > 0) {
+    return `${hours.toString().padStart(2, "0")}:${minutes.toString().padStart(2, "0")}:${seconds.toString().padStart(2, "0")}`;
   }
-  return result;
+  return `${minutes.toString().padStart(2, "0")}:${seconds.toString().padStart(2, "0")}`;
 }
-function convertPiMessage(message) {
+function convertAgentMessage(message, toolTrackers, completedToolResults) {
   if (!message || typeof message !== "object") {
-    return void 0;
+    return { role: "unknown", content: String(message) };
   }
   const msg = message;
-  const role = msg.role;
-  if (typeof role !== "string") {
-    return void 0;
-  }
+  const role = typeof msg.role === "string" ? msg.role : "unknown";
   const content = extractPiTextContent(msg.content);
-  const toolCalls = extractToolCalls4(msg.content);
-  const startTime = typeof msg.timestamp === "number" ? new Date(msg.timestamp).toISOString() : typeof msg.timestamp === "string" ? msg.timestamp : void 0;
+  const toolCalls = extractToolCalls4(msg.content, toolTrackers, completedToolResults);
+  const startTimeVal = typeof msg.timestamp === "number" ? new Date(msg.timestamp).toISOString() : typeof msg.timestamp === "string" ? msg.timestamp : void 0;
+  let msgTokenUsage;
+  if (msg.usage && typeof msg.usage === "object") {
+    const u = msg.usage;
+    const input = toFiniteNumber(u.input);
+    const output = toFiniteNumber(u.output);
+    if (input !== void 0 || output !== void 0) {
+      msgTokenUsage = {
+        input: input ?? 0,
+        output: output ?? 0,
+        ...toFiniteNumber(u.cacheRead) !== void 0 ? { cached: toFiniteNumber(u.cacheRead) } : {}
+      };
+    }
+  }
   const metadata = {};
   if (msg.api) metadata.api = msg.api;
   if (msg.provider) metadata.provider = msg.provider;
   if (msg.model) metadata.model = msg.model;
-  if (msg.usage) metadata.usage = msg.usage;
   if (msg.stopReason) metadata.stopReason = msg.stopReason;
   return {
     role,
     content,
     toolCalls: toolCalls.length > 0 ? toolCalls : void 0,
-    startTime,
-    metadata: Object.keys(metadata).length > 0 ? metadata : void 0
+    startTime: startTimeVal,
+    metadata: Object.keys(metadata).length > 0 ? metadata : void 0,
+    tokenUsage: msgTokenUsage
   };
 }
-function extractToolCalls4(content) {
+function extractToolCalls4(content, toolTrackers, completedToolResults) {
   if (!Array.isArray(content)) {
     return [];
   }
@@ -7677,118 +7959,23 @@ function extractToolCalls4(content) {
       continue;
     }
     const p = part;
-    if (p.type === "tool_use" && typeof p.name === "string") {
-      toolCalls.push({
-        tool: p.name,
-        input: p.input,
-        id: typeof p.id === "string" ? p.id : void 0
-      });
-    }
     if (p.type === "toolCall" && typeof p.name === "string") {
+      const id = typeof p.id === "string" ? p.id : void 0;
+      const tracker = id ? toolTrackers.get(id) : void 0;
+      const completed = id ? completedToolResults.get(id) : void 0;
       toolCalls.push({
         tool: p.name,
         input: p.arguments,
-        id: typeof p.id === "string" ? p.id : void 0
+        id,
+        output: completed?.output,
+        durationMs: completed?.durationMs,
+        startTime: tracker?.startTime,
+        endTime: tracker?.startTime && completed?.durationMs !== void 0 ? new Date(new Date(tracker.startTime).getTime() + completed.durationMs).toISOString() : void 0
       });
     }
-    if (p.type === "tool_result" && typeof p.tool_use_id === "string") {
-      const existing = toolCalls.find((tc) => tc.id === p.tool_use_id);
-      if (existing) {
-        const idx = toolCalls.indexOf(existing);
-        toolCalls[idx] = {
-          ...existing,
-          output: p.content
-        };
-      }
-    }
   }
   return toolCalls;
 }
-function escapeAtSymbols(prompt) {
-  return prompt.replace(/@\[([^\]]+)\]:/g, "[[$1]]:");
-}
-function pickDetail(stderr, stdout) {
-  const errorText = stderr.trim();
-  if (errorText.length > 0) {
-    return errorText;
-  }
-  const stdoutText = stdout.trim();
-  return stdoutText.length > 0 ? stdoutText : void 0;
-}
-function formatTimeoutSuffix3(timeoutMs) {
-  if (!timeoutMs || timeoutMs <= 0) {
-    return "";
-  }
-  const seconds = Math.ceil(timeoutMs / 1e3);
-  return ` after ${seconds}s`;
-}
-async function defaultPiRunner(options) {
-  return await new Promise((resolve, reject) => {
-    const parts = options.executable.split(/\s+/);
-    const executable = parts[0];
-    const executableArgs = parts.slice(1);
-    const allArgs = [...executableArgs, ...options.args];
-    const child = spawn3(executable, allArgs, {
-      cwd: options.cwd,
-      env: options.env,
-      stdio: ["pipe", "pipe", "pipe"],
-      shell: false
-    });
-    let stdout = "";
-    let stderr = "";
-    let timedOut = false;
-    const onAbort = () => {
-      child.kill("SIGTERM");
-    };
-    if (options.signal) {
-      if (options.signal.aborted) {
-        onAbort();
-      } else {
-        options.signal.addEventListener("abort", onAbort, { once: true });
-      }
-    }
-    let timeoutHandle;
-    if (options.timeoutMs && options.timeoutMs > 0) {
-      timeoutHandle = setTimeout(() => {
-        timedOut = true;
-        child.kill("SIGTERM");
-      }, options.timeoutMs);
-      timeoutHandle.unref?.();
-    }
-    child.stdout.setEncoding("utf8");
-    child.stdout.on("data", (chunk) => {
-      stdout += chunk;
-      options.onStdoutChunk?.(chunk);
-    });
-    child.stderr.setEncoding("utf8");
-    child.stderr.on("data", (chunk) => {
-      stderr += chunk;
-      options.onStderrChunk?.(chunk);
-    });
-    child.stdin.end();
-    const cleanup = () => {
-      if (timeoutHandle) {
-        clearTimeout(timeoutHandle);
-      }
-      if (options.signal) {
-        options.signal.removeEventListener("abort", onAbort);
-      }
-    };
-    child.on("error", (error) => {
-      cleanup();
-      reject(error);
-    });
-    child.on("close", (code) => {
-      cleanup();
-      resolve({
-        stdout,
-        stderr,
-        exitCode: typeof code === "number" ? code : -1,
-        timedOut
-      });
-    });
-  });
-}
 
 // src/evaluation/providers/provider-registry.ts
 var ProviderRegistry = class {
@@ -7828,17 +8015,17 @@ var ProviderRegistry = class {
 // src/evaluation/providers/vscode-provider.ts
 import { exec as exec2 } from "node:child_process";
 import { constants as constants3, access as access3, stat as stat4 } from "node:fs/promises";
-import path29 from "node:path";
+import path30 from "node:path";
 import { promisify as promisify3 } from "node:util";
 
 // src/evaluation/providers/vscode/dispatch/agentDispatch.ts
 import { stat as stat3, writeFile as writeFile4 } from "node:fs/promises";
-import path27 from "node:path";
+import path28 from "node:path";
 
 // src/evaluation/providers/vscode/utils/fs.ts
 import { constants as constants2 } from "node:fs";
-import { access as access2, mkdir as mkdir7, readdir, rm as rm2, stat } from "node:fs/promises";
-import path18 from "node:path";
+import { access as access2, mkdir as mkdir8, readdir, rm as rm2, stat } from "node:fs/promises";
+import path19 from "node:path";
 async function pathExists(target) {
   try {
     await access2(target, constants2.F_OK);
@@ -7848,13 +8035,13 @@ async function pathExists(target) {
   }
 }
 async function ensureDir(target) {
-  await mkdir7(target, { recursive: true });
+  await mkdir8(target, { recursive: true });
 }
 async function readDirEntries(target) {
   const entries = await readdir(target, { withFileTypes: true });
   return entries.map((entry) => ({
     name: entry.name,
-    absolutePath: path18.join(target, entry.name),
+    absolutePath: path19.join(target, entry.name),
     isDirectory: entry.isDirectory()
   }));
 }
@@ -7869,9 +8056,9 @@ async function removeIfExists(target) {
 }
 
 // src/evaluation/providers/vscode/utils/path.ts
-import path19 from "node:path";
+import path20 from "node:path";
 function pathToFileUri2(filePath) {
-  const absolutePath = path19.isAbsolute(filePath) ? filePath : path19.resolve(filePath);
+  const absolutePath = path20.isAbsolute(filePath) ? filePath : path20.resolve(filePath);
   const normalizedPath = absolutePath.replace(/\\/g, "/");
   if (/^[a-zA-Z]:\//.test(normalizedPath)) {
     return `file:///${normalizedPath}`;
@@ -7880,7 +8067,7 @@ function pathToFileUri2(filePath) {
 }
 
 // src/evaluation/providers/vscode/dispatch/promptBuilder.ts
-import path20 from "node:path";
+import path21 from "node:path";
 
 // src/evaluation/providers/vscode/utils/template.ts
 function renderTemplate2(content, variables) {
@@ -7972,8 +8159,8 @@ function createBatchRequestPrompt(userQuery, responseFileTmp, responseFileFinal,
   });
 }
 function createBatchOrchestratorPrompt(requestFiles, responseFiles, templateContent) {
-  const requestLines = requestFiles.map((file, index) => `${index + 1}. messages/${path20.basename(file)}`).join("\n");
-  const responseList = responseFiles.map((file) => `"${path20.basename(file)}"`).join(", ");
+  const requestLines = requestFiles.map((file, index) => `${index + 1}. messages/${path21.basename(file)}`).join("\n");
+  const responseList = responseFiles.map((file) => `"${path21.basename(file)}"`).join(", ");
   return renderTemplate2(templateContent, {
     requestFiles: requestLines,
     responseList
@@ -7982,7 +8169,7 @@ function createBatchOrchestratorPrompt(requestFiles, responseFiles, templateCont
 
 // src/evaluation/providers/vscode/dispatch/responseWaiter.ts
 import { readFile as readFile7 } from "node:fs/promises";
-import path21 from "node:path";
+import path22 from "node:path";
 
 // src/evaluation/providers/vscode/utils/time.ts
 function sleep2(ms) {
@@ -8041,7 +8228,7 @@ async function waitForResponseOutput(responseFileFinal, pollInterval = 1e3, sile
 }
 async function waitForBatchResponses(responseFilesFinal, pollInterval = 1e3, silent = false, timeoutMs = DEFAULT_TIMEOUT_MS) {
   if (!silent) {
-    const fileList = responseFilesFinal.map((file) => path21.basename(file)).join(", ");
+    const fileList = responseFilesFinal.map((file) => path22.basename(file)).join(", ");
     console.error(`waiting for ${responseFilesFinal.length} batch response(s): ${fileList}`);
   }
   const deadline = Date.now() + timeoutMs;
@@ -8050,7 +8237,7 @@ async function waitForBatchResponses(responseFilesFinal, pollInterval = 1e3, sil
     while (pending.size > 0) {
       if (Date.now() >= deadline) {
         if (!silent) {
-          const remaining = [...pending].map((f) => path21.basename(f)).join(", ");
+          const remaining = [...pending].map((f) => path22.basename(f)).join(", ");
           console.error(
             `error: timed out after ${Math.round(timeoutMs / 1e3)}s waiting for batch responses. Still pending: ${remaining}`
           );
@@ -8100,16 +8287,16 @@ async function waitForBatchResponses(responseFilesFinal, pollInterval = 1e3, sil
 
 // src/evaluation/providers/vscode/dispatch/vscodeProcess.ts
 import { exec, spawn as spawn4 } from "node:child_process";
-import { mkdir as mkdir8, writeFile as writeFile2 } from "node:fs/promises";
-import path24 from "node:path";
+import { mkdir as mkdir9, writeFile as writeFile2 } from "node:fs/promises";
+import path25 from "node:path";
 import { promisify as promisify2 } from "node:util";
 
 // src/evaluation/providers/vscode/dispatch/constants.ts
-import path23 from "node:path";
+import path24 from "node:path";
 
 // src/paths.ts
 import os2 from "node:os";
-import path22 from "node:path";
+import path23 from "node:path";
 var logged = false;
 function getAgentvHome() {
   const envHome = process.env.AGENTV_HOME;
@@ -8120,19 +8307,19 @@ function getAgentvHome() {
     }
     return envHome;
   }
-  return path22.join(os2.homedir(), ".agentv");
+  return path23.join(os2.homedir(), ".agentv");
 }
 function getWorkspacesRoot() {
-  return path22.join(getAgentvHome(), "workspaces");
+  return path23.join(getAgentvHome(), "workspaces");
 }
 function getSubagentsRoot() {
-  return path22.join(getAgentvHome(), "subagents");
+  return path23.join(getAgentvHome(), "subagents");
 }
 function getTraceStateRoot() {
-  return path22.join(getAgentvHome(), "trace-state");
+  return path23.join(getAgentvHome(), "trace-state");
 }
 function getWorkspacePoolRoot() {
-  return path22.join(getAgentvHome(), "workspace-pool");
+  return path23.join(getAgentvHome(), "workspace-pool");
 }
 
 // src/evaluation/providers/vscode/dispatch/constants.ts
@@ -8140,7 +8327,7 @@ var DEFAULT_LOCK_NAME = "subagent.lock";
 var DEFAULT_ALIVE_FILENAME = ".alive";
 function getDefaultSubagentRoot(vscodeCmd = "code") {
   const folder = vscodeCmd === "code-insiders" ? "vscode-insiders-agents" : "vscode-agents";
-  return path23.join(getSubagentsRoot(), folder);
+  return path24.join(getSubagentsRoot(), folder);
 }
 var DEFAULT_SUBAGENT_ROOT = getDefaultSubagentRoot();
 
@@ -8207,11 +8394,11 @@ async function ensureWorkspaceFocused(workspacePath, workspaceName, subagentDir,
     await raceSpawnError(child);
     return true;
   }
-  const aliveFile = path24.join(subagentDir, DEFAULT_ALIVE_FILENAME);
+  const aliveFile = path25.join(subagentDir, DEFAULT_ALIVE_FILENAME);
   await removeIfExists(aliveFile);
-  const githubAgentsDir = path24.join(subagentDir, ".github", "agents");
-  await mkdir8(githubAgentsDir, { recursive: true });
-  const wakeupDst = path24.join(githubAgentsDir, "wakeup.md");
+  const githubAgentsDir = path25.join(subagentDir, ".github", "agents");
+  await mkdir9(githubAgentsDir, { recursive: true });
+  const wakeupDst = path25.join(githubAgentsDir, "wakeup.md");
   await writeFile2(wakeupDst, DEFAULT_WAKEUP_CONTENT, "utf8");
   const workspaceChild = spawnVsCode(vscodeCmd, [workspacePath], {
     label: "open-workspace"
@@ -8224,7 +8411,7 @@ async function ensureWorkspaceFocused(workspacePath, workspaceName, subagentDir,
     "chat",
     "-m",
     wakeupChatId,
-    `create a file named .alive in the ${path24.basename(subagentDir)} folder`
+    `create a file named .alive in the ${path25.basename(subagentDir)} folder`
   ];
   const wakeupChild = spawnVsCode(vscodeCmd, chatArgs, { label: "send-wakeup-chat" });
   await raceSpawnError(wakeupChild);
@@ -8239,10 +8426,10 @@ async function ensureWorkspaceFocused(workspacePath, workspaceName, subagentDir,
   return true;
 }
 async function launchVsCodeWithChat(subagentDir, chatId, attachmentPaths, requestInstructions, timestamp, vscodeCmd) {
-  const workspacePath = path24.join(subagentDir, `${path24.basename(subagentDir)}.code-workspace`);
-  const messagesDir = path24.join(subagentDir, "messages");
-  await mkdir8(messagesDir, { recursive: true });
-  const reqFile = path24.join(messagesDir, `${timestamp}_req.md`);
+  const workspacePath = path25.join(subagentDir, `${path25.basename(subagentDir)}.code-workspace`);
+  const messagesDir = path25.join(subagentDir, "messages");
+  await mkdir9(messagesDir, { recursive: true });
+  const reqFile = path25.join(messagesDir, `${timestamp}_req.md`);
   await writeFile2(reqFile, requestInstructions, { encoding: "utf8" });
   const reqUri = pathToFileUri2(reqFile);
   const chatArgs = ["-r", "chat", "-m", chatId];
@@ -8250,16 +8437,16 @@ async function launchVsCodeWithChat(subagentDir, chatId, attachmentPaths, reques
     chatArgs.push("-a", attachment);
   }
   chatArgs.push("-a", reqFile);
-  chatArgs.push(`Follow instructions in [${path24.basename(reqFile)}](${reqUri})`);
+  chatArgs.push(`Follow instructions in [${path25.basename(reqFile)}](${reqUri})`);
   const workspaceReady = await ensureWorkspaceFocused(
     workspacePath,
-    path24.basename(subagentDir),
+    path25.basename(subagentDir),
     subagentDir,
     vscodeCmd
   );
   if (!workspaceReady) {
     throw new Error(
-      `VS Code workspace '${path24.basename(subagentDir)}' failed to become ready within the timeout. Check that '${vscodeCmd}' can open workspaces.`
+      `VS Code workspace '${path25.basename(subagentDir)}' failed to become ready within the timeout. Check that '${vscodeCmd}' can open workspaces.`
     );
   }
   await sleep2(500);
@@ -8267,9 +8454,9 @@ async function launchVsCodeWithChat(subagentDir, chatId, attachmentPaths, reques
   await raceSpawnError(child);
 }
 async function launchVsCodeWithBatchChat(subagentDir, chatId, attachmentPaths, chatInstruction, vscodeCmd) {
-  const workspacePath = path24.join(subagentDir, `${path24.basename(subagentDir)}.code-workspace`);
-  const messagesDir = path24.join(subagentDir, "messages");
-  await mkdir8(messagesDir, { recursive: true });
+  const workspacePath = path25.join(subagentDir, `${path25.basename(subagentDir)}.code-workspace`);
+  const messagesDir = path25.join(subagentDir, "messages");
+  await mkdir9(messagesDir, { recursive: true });
   const chatArgs = ["-r", "chat", "-m", chatId];
   for (const attachment of attachmentPaths) {
     chatArgs.push("-a", attachment);
@@ -8277,13 +8464,13 @@ async function launchVsCodeWithBatchChat(subagentDir, chatId, attachmentPaths, c
   chatArgs.push(chatInstruction);
   const workspaceReady = await ensureWorkspaceFocused(
     workspacePath,
-    path24.basename(subagentDir),
+    path25.basename(subagentDir),
     subagentDir,
     vscodeCmd
   );
   if (!workspaceReady) {
     throw new Error(
-      `VS Code workspace '${path24.basename(subagentDir)}' failed to become ready within the timeout. Check that '${vscodeCmd}' can open workspaces.`
+      `VS Code workspace '${path25.basename(subagentDir)}' failed to become ready within the timeout. Check that '${vscodeCmd}' can open workspaces.`
     );
   }
   await sleep2(500);
@@ -8292,11 +8479,11 @@ async function launchVsCodeWithBatchChat(subagentDir, chatId, attachmentPaths, c
 }
 
 // src/evaluation/providers/vscode/dispatch/workspaceManager.ts
-import { copyFile, mkdir as mkdir9, readFile as readFile8, readdir as readdir2, stat as stat2, writeFile as writeFile3 } from "node:fs/promises";
-import path26 from "node:path";
+import { copyFile, mkdir as mkdir10, readFile as readFile8, readdir as readdir2, stat as stat2, writeFile as writeFile3 } from "node:fs/promises";
+import path27 from "node:path";
 
 // src/evaluation/providers/vscode/utils/workspace.ts
-import path25 from "node:path";
+import path26 from "node:path";
 import JSON5 from "json5";
 function transformWorkspacePaths(workspaceContent, templateDir) {
   let workspace;
@@ -8313,10 +8500,10 @@ function transformWorkspacePaths(workspaceContent, templateDir) {
   }
   const transformedFolders = workspace.folders.map((folder) => {
     const folderPath = folder.path;
-    if (path25.isAbsolute(folderPath)) {
+    if (path26.isAbsolute(folderPath)) {
       return folder;
     }
-    const absolutePath = path25.resolve(templateDir, folderPath);
+    const absolutePath = path26.resolve(templateDir, folderPath);
     return {
       ...folder,
       path: absolutePath
@@ -8338,19 +8525,19 @@ function transformWorkspacePaths(workspaceContent, templateDir) {
       if (locationMap && typeof locationMap === "object") {
         const transformedMap = {};
         for (const [locationPath, value] of Object.entries(locationMap)) {
-          const isAbsolute = path25.isAbsolute(locationPath);
+          const isAbsolute = path26.isAbsolute(locationPath);
           if (isAbsolute) {
             transformedMap[locationPath] = value;
           } else {
             const firstGlobIndex = locationPath.search(/[*]/);
             if (firstGlobIndex === -1) {
-              const resolvedPath = path25.resolve(templateDir, locationPath).replace(/\\/g, "/");
+              const resolvedPath = path26.resolve(templateDir, locationPath).replace(/\\/g, "/");
               transformedMap[resolvedPath] = value;
             } else {
               const basePathEnd = locationPath.lastIndexOf("/", firstGlobIndex);
               const basePath = basePathEnd !== -1 ? locationPath.substring(0, basePathEnd) : ".";
               const patternPath = locationPath.substring(basePathEnd !== -1 ? basePathEnd : 0);
-              const resolvedPath = (path25.resolve(templateDir, basePath) + patternPath).replace(
+              const resolvedPath = (path26.resolve(templateDir, basePath) + patternPath).replace(
                 /\\/g,
                 "/"
               );
@@ -8391,7 +8578,7 @@ async function findUnlockedSubagent(subagentRoot) {
     number: Number.parseInt(entry.name.split("-")[1] ?? "", 10)
   })).filter((entry) => Number.isInteger(entry.number)).sort((a, b) => a.number - b.number);
   for (const subagent of subagents) {
-    const lockFile = path26.join(subagent.absolutePath, DEFAULT_LOCK_NAME);
+    const lockFile = path27.join(subagent.absolutePath, DEFAULT_LOCK_NAME);
     if (!await pathExists(lockFile)) {
       return subagent.absolutePath;
     }
@@ -8401,7 +8588,7 @@ async function findUnlockedSubagent(subagentRoot) {
 async function copyAgentConfig(subagentDir, workspaceTemplate, cwd) {
   let workspaceContent;
   if (workspaceTemplate) {
-    const workspaceSrc = path26.resolve(workspaceTemplate);
+    const workspaceSrc = path27.resolve(workspaceTemplate);
     if (!await pathExists(workspaceSrc)) {
       throw new Error(`workspace template not found: ${workspaceSrc}`);
     }
@@ -8414,13 +8601,13 @@ async function copyAgentConfig(subagentDir, workspaceTemplate, cwd) {
   } else {
     workspaceContent = DEFAULT_WORKSPACE_TEMPLATE;
   }
-  const workspaceName = `${path26.basename(subagentDir)}.code-workspace`;
-  const workspaceDst = path26.join(subagentDir, workspaceName);
-  const templateDir = workspaceTemplate ? path26.dirname(path26.resolve(workspaceTemplate)) : subagentDir;
+  const workspaceName = `${path27.basename(subagentDir)}.code-workspace`;
+  const workspaceDst = path27.join(subagentDir, workspaceName);
+  const templateDir = workspaceTemplate ? path27.dirname(path27.resolve(workspaceTemplate)) : subagentDir;
   const workspaceJson = JSON.stringify(workspaceContent, null, 2);
   let transformedContent = transformWorkspacePaths(workspaceJson, templateDir);
   if (cwd) {
-    const absCwd = path26.resolve(cwd);
+    const absCwd = path27.resolve(cwd);
     const parsed = JSON.parse(transformedContent);
     const alreadyPresent = parsed.folders.some((f) => f.path === absCwd);
     if (!alreadyPresent) {
@@ -8429,35 +8616,35 @@ async function copyAgentConfig(subagentDir, workspaceTemplate, cwd) {
     }
   }
   await writeFile3(workspaceDst, transformedContent, "utf8");
-  const messagesDir = path26.join(subagentDir, "messages");
-  await mkdir9(messagesDir, { recursive: true });
+  const messagesDir = path27.join(subagentDir, "messages");
+  await mkdir10(messagesDir, { recursive: true });
   return { workspace: workspaceDst, messagesDir };
 }
 async function createSubagentLock(subagentDir) {
-  const messagesDir = path26.join(subagentDir, "messages");
+  const messagesDir = path27.join(subagentDir, "messages");
   if (await pathExists(messagesDir)) {
     const files = await readdir2(messagesDir);
     await Promise.all(
       files.map(async (file) => {
-        const target = path26.join(messagesDir, file);
+        const target = path27.join(messagesDir, file);
         await removeIfExists(target);
       })
     );
   }
-  const githubAgentsDir = path26.join(subagentDir, ".github", "agents");
+  const githubAgentsDir = path27.join(subagentDir, ".github", "agents");
   if (await pathExists(githubAgentsDir)) {
     const agentFiles = await readdir2(githubAgentsDir);
     const preservedFiles = /* @__PURE__ */ new Set(["wakeup.md", "subagent.md"]);
     await Promise.all(
-      agentFiles.filter((file) => file.endsWith(".md") && !preservedFiles.has(file)).map((file) => removeIfExists(path26.join(githubAgentsDir, file)))
+      agentFiles.filter((file) => file.endsWith(".md") && !preservedFiles.has(file)).map((file) => removeIfExists(path27.join(githubAgentsDir, file)))
     );
   }
-  const lockFile = path26.join(subagentDir, DEFAULT_LOCK_NAME);
+  const lockFile = path27.join(subagentDir, DEFAULT_LOCK_NAME);
   await writeFile3(lockFile, "", { encoding: "utf8" });
   return lockFile;
 }
 async function removeSubagentLock(subagentDir) {
-  const lockFile = path26.join(subagentDir, DEFAULT_LOCK_NAME);
+  const lockFile = path27.join(subagentDir, DEFAULT_LOCK_NAME);
   await removeIfExists(lockFile);
 }
 async function prepareSubagentDirectory(subagentDir, promptFile, chatId, workspaceTemplate, dryRun, cwd) {
@@ -8477,9 +8664,9 @@ async function prepareSubagentDirectory(subagentDir, promptFile, chatId, workspa
     return 1;
   }
   if (promptFile) {
-    const githubAgentsDir = path26.join(subagentDir, ".github", "agents");
-    await mkdir9(githubAgentsDir, { recursive: true });
-    const agentFile = path26.join(githubAgentsDir, `${chatId}.md`);
+    const githubAgentsDir = path27.join(subagentDir, ".github", "agents");
+    await mkdir10(githubAgentsDir, { recursive: true });
+    const agentFile = path27.join(githubAgentsDir, `${chatId}.md`);
     try {
       await copyFile(promptFile, agentFile);
     } catch (error) {
@@ -8498,7 +8685,7 @@ async function resolvePromptFile(promptFile) {
   if (!promptFile) {
     return void 0;
   }
-  const resolvedPrompt = path27.resolve(promptFile);
+  const resolvedPrompt = path28.resolve(promptFile);
   if (!await pathExists(resolvedPrompt)) {
     throw new Error(`Prompt file not found: ${resolvedPrompt}`);
   }
@@ -8514,7 +8701,7 @@ async function resolveAttachments(extraAttachments) {
   }
   const resolved = [];
   for (const attachment of extraAttachments) {
-    const resolvedPath = path27.resolve(attachment);
+    const resolvedPath = path28.resolve(attachment);
     if (!await pathExists(resolvedPath)) {
       throw new Error(`Attachment not found: ${resolvedPath}`);
     }
@@ -8556,7 +8743,7 @@ async function dispatchAgentSession(options) {
         error: "No unlocked subagents available. Provision additional subagents with: subagent code provision --subagents <desired_total>"
       };
     }
-    const subagentName = path27.basename(subagentDir);
+    const subagentName = path28.basename(subagentDir);
     const chatId = Math.random().toString(16).slice(2, 10);
     const preparationResult = await prepareSubagentDirectory(
       subagentDir,
@@ -8584,9 +8771,9 @@ async function dispatchAgentSession(options) {
       };
     }
     const timestamp = generateTimestamp();
-    const messagesDir = path27.join(subagentDir, "messages");
-    const responseFileTmp = path27.join(messagesDir, `${timestamp}_res.tmp.md`);
-    const responseFileFinal = path27.join(messagesDir, `${timestamp}_res.md`);
+    const messagesDir = path28.join(subagentDir, "messages");
+    const responseFileTmp = path28.join(messagesDir, `${timestamp}_res.tmp.md`);
+    const responseFileFinal = path28.join(messagesDir, `${timestamp}_res.md`);
     const requestInstructions = createRequestPrompt(
       userQuery,
       responseFileTmp,
@@ -8691,7 +8878,7 @@ async function dispatchBatchAgent(options) {
         error: "No unlocked subagents available. Provision additional subagents with: subagent code provision --subagents <desired_total>"
       };
     }
-    subagentName = path27.basename(subagentDir);
+    subagentName = path28.basename(subagentDir);
     const chatId = Math.random().toString(16).slice(2, 10);
     const preparationResult = await prepareSubagentDirectory(
       subagentDir,
@@ -8722,17 +8909,17 @@ async function dispatchBatchAgent(options) {
       };
     }
     const timestamp = generateTimestamp();
-    const messagesDir = path27.join(subagentDir, "messages");
+    const messagesDir = path28.join(subagentDir, "messages");
     requestFiles = userQueries.map(
-      (_, index) => path27.join(messagesDir, `${timestamp}_${index}_req.md`)
+      (_, index) => path28.join(messagesDir, `${timestamp}_${index}_req.md`)
     );
     const responseTmpFiles = userQueries.map(
-      (_, index) => path27.join(messagesDir, `${timestamp}_${index}_res.tmp.md`)
+      (_, index) => path28.join(messagesDir, `${timestamp}_${index}_res.tmp.md`)
     );
     responseFilesFinal = userQueries.map(
-      (_, index) => path27.join(messagesDir, `${timestamp}_${index}_res.md`)
+      (_, index) => path28.join(messagesDir, `${timestamp}_${index}_res.md`)
     );
-    const orchestratorFile = path27.join(messagesDir, `${timestamp}_orchestrator.md`);
+    const orchestratorFile = path28.join(messagesDir, `${timestamp}_orchestrator.md`);
     if (!dryRun) {
       await Promise.all(
         userQueries.map((query, index) => {
@@ -8818,7 +9005,7 @@ async function dispatchBatchAgent(options) {
 
 // src/evaluation/providers/vscode/dispatch/provision.ts
 import { writeFile as writeFile5 } from "node:fs/promises";
-import path28 from "node:path";
+import path29 from "node:path";
 var DEFAULT_WORKSPACE_TEMPLATE2 = {
   folders: [
     {
@@ -8849,7 +9036,7 @@ async function provisionSubagents(options) {
   if (!Number.isInteger(subagents) || subagents < 1) {
     throw new Error("subagents must be a positive integer");
   }
-  const targetPath = path28.resolve(targetRoot);
+  const targetPath = path29.resolve(targetRoot);
   if (!dryRun) {
     await ensureDir(targetPath);
   }
@@ -8869,7 +9056,7 @@ async function provisionSubagents(options) {
         continue;
       }
       highestNumber = Math.max(highestNumber, parsed);
-      const lockFile = path28.join(entry.absolutePath, lockName);
+      const lockFile = path29.join(entry.absolutePath, lockName);
       const locked = await pathExists(lockFile);
       if (locked) {
         lockedSubagents.add(entry.absolutePath);
@@ -8886,10 +9073,10 @@ async function provisionSubagents(options) {
       break;
     }
     const subagentDir = subagent.absolutePath;
-    const githubAgentsDir = path28.join(subagentDir, ".github", "agents");
-    const lockFile = path28.join(subagentDir, lockName);
-    const workspaceDst = path28.join(subagentDir, `${path28.basename(subagentDir)}.code-workspace`);
-    const wakeupDst = path28.join(githubAgentsDir, "wakeup.md");
+    const githubAgentsDir = path29.join(subagentDir, ".github", "agents");
+    const lockFile = path29.join(subagentDir, lockName);
+    const workspaceDst = path29.join(subagentDir, `${path29.basename(subagentDir)}.code-workspace`);
+    const wakeupDst = path29.join(githubAgentsDir, "wakeup.md");
     const isLocked = await pathExists(lockFile);
     if (isLocked && !force) {
       continue;
@@ -8927,10 +9114,10 @@ async function provisionSubagents(options) {
   let nextIndex = highestNumber;
   while (subagentsProvisioned < subagents) {
     nextIndex += 1;
-    const subagentDir = path28.join(targetPath, `subagent-${nextIndex}`);
-    const githubAgentsDir = path28.join(subagentDir, ".github", "agents");
-    const workspaceDst = path28.join(subagentDir, `${path28.basename(subagentDir)}.code-workspace`);
-    const wakeupDst = path28.join(githubAgentsDir, "wakeup.md");
+    const subagentDir = path29.join(targetPath, `subagent-${nextIndex}`);
+    const githubAgentsDir = path29.join(subagentDir, ".github", "agents");
+    const workspaceDst = path29.join(subagentDir, `${path29.basename(subagentDir)}.code-workspace`);
+    const wakeupDst = path29.join(githubAgentsDir, "wakeup.md");
     if (!dryRun) {
       await ensureDir(subagentDir);
       await ensureDir(githubAgentsDir);
@@ -9120,7 +9307,7 @@ var VSCodeProvider = class {
 async function locateVSCodeExecutable(candidate) {
   const includesPathSeparator = candidate.includes("/") || candidate.includes("\\");
   if (includesPathSeparator) {
-    const resolved = path29.isAbsolute(candidate) ? candidate : path29.resolve(candidate);
+    const resolved = path30.isAbsolute(candidate) ? candidate : path30.resolve(candidate);
     try {
       await access3(resolved, constants3.F_OK);
       return resolved;
@@ -9149,7 +9336,7 @@ async function resolveWorkspaceTemplateFile(template) {
     return void 0;
   }
   try {
-    const stats = await stat4(path29.resolve(template));
+    const stats = await stat4(path30.resolve(template));
     return stats.isFile() ? template : void 0;
   } catch {
     return template;
@@ -9173,7 +9360,7 @@ function buildMandatoryPrereadBlock2(attachmentFiles) {
     return "";
   }
   const buildList = (files) => files.map((absolutePath) => {
-    const fileName = path29.basename(absolutePath);
+    const fileName = path30.basename(absolutePath);
     const fileUri = pathToFileUri3(absolutePath);
     return `* [${fileName}](${fileUri})`;
   });
@@ -9194,7 +9381,7 @@ function collectAttachmentFiles(attachments) {
   }
   const unique = /* @__PURE__ */ new Map();
   for (const attachment of attachments) {
-    const absolutePath = path29.resolve(attachment);
+    const absolutePath = path30.resolve(attachment);
     if (!unique.has(absolutePath)) {
       unique.set(absolutePath, absolutePath);
     }
@@ -9202,7 +9389,7 @@ function collectAttachmentFiles(attachments) {
   return Array.from(unique.values());
 }
 function pathToFileUri3(filePath) {
-  const absolutePath = path29.isAbsolute(filePath) ? filePath : path29.resolve(filePath);
+  const absolutePath = path30.isAbsolute(filePath) ? filePath : path30.resolve(filePath);
   const normalizedPath = absolutePath.replace(/\\/g, "/");
   if (/^[a-zA-Z]:\//.test(normalizedPath)) {
     return `file:///${normalizedPath}`;
@@ -9215,7 +9402,7 @@ function normalizeAttachments(attachments) {
   }
   const deduped = /* @__PURE__ */ new Set();
   for (const attachment of attachments) {
-    deduped.add(path29.resolve(attachment));
+    deduped.add(path30.resolve(attachment));
   }
   return Array.from(deduped);
 }
@@ -9224,7 +9411,7 @@ function mergeAttachments(all) {
   for (const list of all) {
     if (!list) continue;
     for (const inputFile of list) {
-      deduped.add(path29.resolve(inputFile));
+      deduped.add(path30.resolve(inputFile));
     }
   }
   return deduped.size > 0 ? Array.from(deduped) : void 0;
@@ -9273,7 +9460,7 @@ total unlocked subagents available: ${result.created.length + result.skippedExis
 // src/evaluation/providers/targets-file.ts
 import { constants as constants4 } from "node:fs";
 import { access as access4, readFile as readFile9 } from "node:fs/promises";
-import path30 from "node:path";
+import path31 from "node:path";
 import { parse as parse4 } from "yaml";
 function isRecord(value) {
   return typeof value === "object" && value !== null && !Array.isArray(value);
@@ -9310,7 +9497,7 @@ async function fileExists3(filePath) {
   }
 }
 async function readTargetDefinitions(filePath) {
-  const absolutePath = path30.resolve(filePath);
+  const absolutePath = path31.resolve(filePath);
   if (!await fileExists3(absolutePath)) {
     throw new Error(`targets.yaml not found at ${absolutePath}`);
   }
@@ -9330,16 +9517,16 @@ function listTargetNames(definitions) {
 }
 
 // src/evaluation/providers/provider-discovery.ts
-import path31 from "node:path";
+import path32 from "node:path";
 import fg from "fast-glob";
 async function discoverProviders(registry, baseDir) {
   const patterns = ["*.ts", "*.js", "*.mts", "*.mjs"];
   const candidateDirs = [];
-  let dir = path31.resolve(baseDir);
-  const root = path31.parse(dir).root;
+  let dir = path32.resolve(baseDir);
+  const root = path32.parse(dir).root;
   while (dir !== root) {
-    candidateDirs.push(path31.join(dir, ".agentv", "providers"));
-    dir = path31.dirname(dir);
+    candidateDirs.push(path32.join(dir, ".agentv", "providers"));
+    dir = path32.dirname(dir);
   }
   let files = [];
   for (const providersDir of candidateDirs) {
@@ -9355,7 +9542,7 @@ async function discoverProviders(registry, baseDir) {
   }
   const discoveredKinds = [];
   for (const filePath of files) {
-    const basename = path31.basename(filePath);
+    const basename = path32.basename(filePath);
     const kindName = basename.replace(/\.(ts|js|mts|mjs)$/, "");
     if (registry.has(kindName)) {
       continue;
@@ -9373,7 +9560,7 @@ async function discoverProviders(registry, baseDir) {
 // src/evaluation/providers/index.ts
 function createBuiltinProviderRegistry() {
   const registry = new ProviderRegistry();
-  registry.register("openai", (t) => new OpenAIProvider(t.name, t.config)).register("openrouter", (t) => new OpenRouterProvider(t.name, t.config)).register("azure", (t) => new AzureProvider(t.name, t.config)).register("anthropic", (t) => new AnthropicProvider(t.name, t.config)).register("gemini", (t) => new GeminiProvider(t.name, t.config)).register("cli", (t) => new CliProvider(t.name, t.config)).register("codex", (t) => new CodexProvider(t.name, t.config)).register("copilot-sdk", (t) => new CopilotSdkProvider(t.name, t.config)).register("copilot-cli", (t) => new CopilotCliProvider(t.name, t.config)).register("pi-coding-agent", (t) => new PiCodingAgentProvider(t.name, t.config)).register("pi-agent-sdk", (t) => new PiAgentSdkProvider(t.name, t.config)).register("claude-cli", (t) => new ClaudeCliProvider(t.name, t.config)).register("claude", (t) => new ClaudeCliProvider(t.name, t.config)).register("claude-sdk", (t) => new ClaudeSdkProvider(t.name, t.config)).register("mock", (t) => new MockProvider(t.name, t.config)).register("agentv", (t) => new AgentvProvider(t.name, t.config)).register("vscode", (t) => new VSCodeProvider(t.name, t.config, "vscode")).register(
+  registry.register("openai", (t) => new OpenAIProvider(t.name, t.config)).register("openrouter", (t) => new OpenRouterProvider(t.name, t.config)).register("azure", (t) => new AzureProvider(t.name, t.config)).register("anthropic", (t) => new AnthropicProvider(t.name, t.config)).register("gemini", (t) => new GeminiProvider(t.name, t.config)).register("cli", (t) => new CliProvider(t.name, t.config)).register("codex", (t) => new CodexProvider(t.name, t.config)).register("copilot-sdk", (t) => new CopilotSdkProvider(t.name, t.config)).register("copilot-cli", (t) => new CopilotCliProvider(t.name, t.config)).register("pi-coding-agent", (t) => new PiCodingAgentProvider(t.name, t.config)).register("pi-cli", (t) => new PiCliProvider(t.name, t.config)).register("claude-cli", (t) => new ClaudeCliProvider(t.name, t.config)).register("claude", (t) => new ClaudeCliProvider(t.name, t.config)).register("claude-sdk", (t) => new ClaudeSdkProvider(t.name, t.config)).register("mock", (t) => new MockProvider(t.name, t.config)).register("agentv", (t) => new AgentvProvider(t.name, t.config)).register("vscode", (t) => new VSCodeProvider(t.name, t.config, "vscode")).register(
     "vscode-insiders",
     (t) => new VSCodeProvider(t.name, t.config, "vscode-insiders")
   );
@@ -9564,15 +9751,15 @@ async function execFileWithStdinNode(argv, stdinPayload, options) {
   });
 }
 async function execShellWithStdin(command, stdinPayload, options = {}) {
-  const { mkdir: mkdir15, readFile: readFile12, rm: rm6, writeFile: writeFile9 } = await import("node:fs/promises");
+  const { mkdir: mkdir16, readFile: readFile12, rm: rm6, writeFile: writeFile9 } = await import("node:fs/promises");
   const { tmpdir: tmpdir3 } = await import("node:os");
-  const path44 = await import("node:path");
-  const { randomUUID: randomUUID9 } = await import("node:crypto");
-  const dir = path44.join(tmpdir3(), `agentv-exec-${randomUUID9()}`);
-  await mkdir15(dir, { recursive: true });
-  const stdinPath = path44.join(dir, "stdin.txt");
-  const stdoutPath = path44.join(dir, "stdout.txt");
-  const stderrPath = path44.join(dir, "stderr.txt");
+  const path45 = await import("node:path");
+  const { randomUUID: randomUUID10 } = await import("node:crypto");
+  const dir = path45.join(tmpdir3(), `agentv-exec-${randomUUID10()}`);
+  await mkdir16(dir, { recursive: true });
+  const stdinPath = path45.join(dir, "stdin.txt");
+  const stdoutPath = path45.join(dir, "stdout.txt");
+  const stderrPath = path45.join(dir, "stderr.txt");
   await writeFile9(stdinPath, stdinPayload, "utf8");
   const wrappedCommand = process.platform === "win32" ? `(${command}) < ${shellEscapePath(stdinPath)} > ${shellEscapePath(stdoutPath)} 2> ${shellEscapePath(stderrPath)}` : `(${command}) < ${shellEscapePath(stdinPath)} > ${shellEscapePath(stdoutPath)} 2> ${shellEscapePath(stderrPath)}`;
   const { spawn: spawn5 } = await import("node:child_process");
@@ -10051,7 +10238,7 @@ import { generateText as generateText3 } from "ai";
 
 // src/evaluation/evaluators/llm-grader.ts
 import fs2 from "node:fs/promises";
-import path32 from "node:path";
+import path33 from "node:path";
 import { generateText as generateText2, stepCountIs, tool } from "ai";
 import { z as z3 } from "zod";
 var DEFAULT_MAX_STEPS = 10;
@@ -10240,7 +10427,7 @@ ${context.fileChanges}`;
   async evaluateWithRubrics(context, graderProvider, rubrics) {
     if (!rubrics || rubrics.length === 0) {
       throw new Error(
-        `No rubrics found for evaluator "${context.evaluator?.name ?? "llm-grader"}". Run "agentv generate rubrics" first.`
+        `No rubrics found for evaluator "${context.evaluator?.name ?? "llm-grader"}". Add rubric criteria under assertions or use the agentv-eval-writer skill for authoring help.`
       );
     }
     const hasScoreRanges = rubrics.some((r) => r.score_ranges && r.score_ranges.length > 0);
@@ -10906,8 +11093,8 @@ function calculateScoreRangeResult(result, rubrics) {
   };
 }
 function resolveSandboxed(basePath, relativePath) {
-  const resolved = path32.resolve(basePath, relativePath);
-  if (!resolved.startsWith(basePath + path32.sep) && resolved !== basePath) {
+  const resolved = path33.resolve(basePath, relativePath);
+  if (!resolved.startsWith(basePath + path33.sep) && resolved !== basePath) {
     throw new Error(`Path '${relativePath}' is outside the workspace`);
   }
   return resolved;
@@ -10997,11 +11184,11 @@ async function searchDirectory(dirPath, workspacePath, regex, matches) {
   for (const entry of entries) {
     if (matches.length >= MAX_SEARCH_MATCHES) return;
     if (SEARCH_SKIP_DIRS.has(entry.name)) continue;
-    const fullPath = path32.join(dirPath, entry.name);
+    const fullPath = path33.join(dirPath, entry.name);
     if (entry.isDirectory()) {
       await searchDirectory(fullPath, workspacePath, regex, matches);
     } else if (entry.isFile()) {
-      const ext = path32.extname(entry.name).toLowerCase();
+      const ext = path33.extname(entry.name).toLowerCase();
       if (BINARY_EXTENSIONS.has(ext)) continue;
       try {
         const stat8 = await fs2.stat(fullPath);
@@ -11013,7 +11200,7 @@ async function searchDirectory(dirPath, workspacePath, regex, matches) {
           regex.lastIndex = 0;
           if (regex.test(lines[i])) {
             matches.push({
-              file: path32.relative(workspacePath, fullPath),
+              file: path33.relative(workspacePath, fullPath),
               line: i + 1,
               text: lines[i].substring(0, 200)
             });
@@ -11648,115 +11835,115 @@ var FieldAccuracyEvaluator = class {
    * Evaluate a single field against the expected value.
    */
   evaluateField(fieldConfig, candidateData, expectedData) {
-    const { path: path44, match, required = true, weight = 1 } = fieldConfig;
-    const candidateValue = resolvePath(candidateData, path44);
-    const expectedValue = resolvePath(expectedData, path44);
+    const { path: path45, match, required = true, weight = 1 } = fieldConfig;
+    const candidateValue = resolvePath(candidateData, path45);
+    const expectedValue = resolvePath(expectedData, path45);
     if (expectedValue === void 0) {
       return {
-        path: path44,
+        path: path45,
         score: 1,
         // No expected value means no comparison needed
         weight,
         hit: true,
-        message: `${path44}: no expected value`
+        message: `${path45}: no expected value`
       };
     }
     if (candidateValue === void 0) {
       if (required) {
         return {
-          path: path44,
+          path: path45,
           score: 0,
           weight,
           hit: false,
-          message: `${path44} (required, missing)`
+          message: `${path45} (required, missing)`
         };
       }
       return {
-        path: path44,
+        path: path45,
         score: 1,
         // Don't penalize missing optional fields
         weight: 0,
         // Zero weight means it won't affect the score
         hit: true,
-        message: `${path44}: optional field missing`
+        message: `${path45}: optional field missing`
       };
     }
     switch (match) {
       case "exact":
-        return this.compareExact(path44, candidateValue, expectedValue, weight);
+        return this.compareExact(path45, candidateValue, expectedValue, weight);
       case "numeric_tolerance":
         return this.compareNumericTolerance(
-          path44,
+          path45,
           candidateValue,
           expectedValue,
           fieldConfig,
           weight
         );
       case "date":
-        return this.compareDate(path44, candidateValue, expectedValue, fieldConfig, weight);
+        return this.compareDate(path45, candidateValue, expectedValue, fieldConfig, weight);
       default:
         return {
-          path: path44,
+          path: path45,
           score: 0,
           weight,
           hit: false,
-          message: `${path44}: unknown match type "${match}"`
+          message: `${path45}: unknown match type "${match}"`
         };
     }
   }
   /**
    * Exact equality comparison.
    */
-  compareExact(path44, candidateValue, expectedValue, weight) {
+  compareExact(path45, candidateValue, expectedValue, weight) {
     if (deepEqual(candidateValue, expectedValue)) {
       return {
-        path: path44,
+        path: path45,
         score: 1,
         weight,
         hit: true,
-        message: path44
+        message: path45
       };
     }
     if (typeof candidateValue !== typeof expectedValue) {
       return {
-        path: path44,
+        path: path45,
         score: 0,
         weight,
         hit: false,
-        message: `${path44} (type mismatch: got ${typeof candidateValue}, expected ${typeof expectedValue})`
+        message: `${path45} (type mismatch: got ${typeof candidateValue}, expected ${typeof expectedValue})`
       };
     }
     return {
-      path: path44,
+      path: path45,
       score: 0,
       weight,
       hit: false,
-      message: `${path44} (value mismatch)`
+      message: `${path45} (value mismatch)`
     };
   }
   /**
    * Numeric comparison with absolute or relative tolerance.
    */
-  compareNumericTolerance(path44, candidateValue, expectedValue, fieldConfig, weight) {
+  compareNumericTolerance(path45, candidateValue, expectedValue, fieldConfig, weight) {
     const { tolerance = 0, relative = false } = fieldConfig;
     const candidateNum = toNumber(candidateValue);
     const expectedNum = toNumber(expectedValue);
     if (candidateNum === null || expectedNum === null) {
       return {
-        path: path44,
+        path: path45,
         score: 0,
         weight,
         hit: false,
-        message: `${path44} (non-numeric value)`
+        message: `${path45} (non-numeric value)`
       };
     }
     if (!Number.isFinite(candidateNum) || !Number.isFinite(expectedNum)) {
       return {
-        path: path44,
+        path: path45,
         score: 0,
         weight,
         hit: false,
-        message: `${path44} (invalid numeric value)`
+        message: `${path45} (invalid numeric value)`
       };
     }
     const diff = Math.abs(candidateNum - expectedNum);
@@ -11769,61 +11956,61 @@ var FieldAccuracyEvaluator = class {
     }
     if (withinTolerance) {
       return {
-        path: path44,
+        path: path45,
         score: 1,
         weight,
         hit: true,
-        message: `${path44} (within tolerance: diff=${diff.toFixed(2)})`
+        message: `${path45} (within tolerance: diff=${diff.toFixed(2)})`
       };
     }
     return {
-      path: path44,
+      path: path45,
       score: 0,
       weight,
       hit: false,
-      message: `${path44} (outside tolerance: diff=${diff.toFixed(2)}, tolerance=${tolerance})`
+      message: `${path45} (outside tolerance: diff=${diff.toFixed(2)}, tolerance=${tolerance})`
     };
   }
   /**
    * Date comparison with format normalization.
    */
-  compareDate(path44, candidateValue, expectedValue, fieldConfig, weight) {
+  compareDate(path45, candidateValue, expectedValue, fieldConfig, weight) {
     const formats = fieldConfig.formats ?? DEFAULT_DATE_FORMATS;
     const candidateDate = parseDate(String(candidateValue), formats);
     const expectedDate = parseDate(String(expectedValue), formats);
     if (candidateDate === null) {
       return {
-        path: path44,
+        path: path45,
         score: 0,
         weight,
         hit: false,
-        message: `${path44} (unparseable candidate date)`
+        message: `${path45} (unparseable candidate date)`
       };
     }
     if (expectedDate === null) {
       return {
-        path: path44,
+        path: path45,
         score: 0,
         weight,
         hit: false,
-        message: `${path44} (unparseable expected date)`
+        message: `${path45} (unparseable expected date)`
       };
     }
     if (candidateDate.getFullYear() === expectedDate.getFullYear() && candidateDate.getMonth() === expectedDate.getMonth() && candidateDate.getDate() === expectedDate.getDate()) {
       return {
-        path: path44,
+        path: path45,
         score: 1,
         weight,
         hit: true,
-        message: path44
+        message: path45
       };
     }
     return {
-      path: path44,
+      path: path45,
       score: 0,
       weight,
       hit: false,
-      message: `${path44} (date mismatch: got ${formatDateISO(candidateDate)}, expected ${formatDateISO(expectedDate)})`
+      message: `${path45} (date mismatch: got ${formatDateISO(candidateDate)}, expected ${formatDateISO(expectedDate)})`
     };
   }
   /**
@@ -11856,11 +12043,11 @@ var FieldAccuracyEvaluator = class {
     };
   }
 };
-function resolvePath(obj, path44) {
-  if (!path44 || !obj) {
+function resolvePath(obj, path45) {
+  if (!path45 || !obj) {
     return void 0;
   }
-  const parts = path44.split(/\.|\[|\]/).filter((p) => p.length > 0);
+  const parts = path45.split(/\.|\[|\]/).filter((p) => p.length > 0);
   let current = obj;
   for (const part of parts) {
     if (current === null || current === void 0) {
@@ -12026,9 +12213,7 @@ var PROVIDER_TOOL_SEMANTICS = {
   "claude-sdk": CLAUDE_MATCHER,
   codex: CODEX_MATCHER,
   "pi-coding-agent": PI_CODING_AGENT_MATCHER,
-  // pi-agent-sdk has no tools, so skill detection is a no-op. Kept for completeness.
-  // TODO: consider removing pi-agent-sdk provider entirely.
-  "pi-agent-sdk": PI_CODING_AGENT_MATCHER,
+  "pi-cli": PI_CODING_AGENT_MATCHER,
   "copilot-cli": COPILOT_MATCHER,
   "copilot-sdk": COPILOT_MATCHER,
   vscode: COPILOT_MATCHER,
@@ -12343,8 +12528,8 @@ var TokenUsageEvaluator = class {
 };
 
 // src/evaluation/evaluators/tool-trajectory.ts
-function getNestedValue(obj, path44) {
-  const parts = path44.split(".");
+function getNestedValue(obj, path45) {
+  const parts = path45.split(".");
   let current = obj;
   for (const part of parts) {
     if (current === null || current === void 0 || typeof current !== "object") {
@@ -12964,9 +13149,9 @@ function runEqualsAssertion(output, value) {
 }
 
 // src/evaluation/orchestrator.ts
-import { createHash as createHash2, randomUUID as randomUUID8 } from "node:crypto";
-import { copyFile as copyFile2, mkdir as mkdir13, readdir as readdir6, stat as stat7 } from "node:fs/promises";
-import path41 from "node:path";
+import { createHash as createHash2, randomUUID as randomUUID9 } from "node:crypto";
+import { copyFile as copyFile2, mkdir as mkdir14, readdir as readdir6, stat as stat7 } from "node:fs/promises";
+import path42 from "node:path";
 import micromatch3 from "micromatch";
 
 // ../../node_modules/.bun/yocto-queue@1.2.2/node_modules/yocto-queue/index.js
@@ -13180,7 +13365,7 @@ var InlineAssertEvaluator = class {
 };
 
 // src/evaluation/evaluators/prompt-resolution.ts
-import path33 from "node:path";
+import path34 from "node:path";
 async function resolveCustomPrompt(promptConfig, context, timeoutMs) {
   if (promptConfig.resolvedPromptScript && promptConfig.resolvedPromptScript.length > 0) {
     if (!context) {
@@ -13226,7 +13411,7 @@ async function executePromptTemplate(script, context, config, timeoutMs) {
   };
   const inputJson = JSON.stringify(toSnakeCaseDeep(payload), null, 2);
   const scriptPath = script[script.length - 1];
-  const cwd = path33.dirname(scriptPath);
+  const cwd = path34.dirname(scriptPath);
   try {
     const stdout = await executeScript(script, inputJson, timeoutMs, cwd);
     const prompt = stdout.trim();
@@ -13498,16 +13683,16 @@ function createBuiltinRegistry() {
 }
 
 // src/evaluation/registry/assertion-discovery.ts
-import path34 from "node:path";
+import path35 from "node:path";
 import fg2 from "fast-glob";
 async function discoverAssertions(registry, baseDir) {
   const patterns = ["*.ts", "*.js", "*.mts", "*.mjs"];
   const candidateDirs = [];
-  let dir = path34.resolve(baseDir);
-  const root = path34.parse(dir).root;
+  let dir = path35.resolve(baseDir);
+  const root = path35.parse(dir).root;
   while (dir !== root) {
-    candidateDirs.push(path34.join(dir, ".agentv", "assertions"));
-    dir = path34.dirname(dir);
+    candidateDirs.push(path35.join(dir, ".agentv", "assertions"));
+    dir = path35.dirname(dir);
   }
   let files = [];
   for (const assertionsDir of candidateDirs) {
@@ -13523,7 +13708,7 @@ async function discoverAssertions(registry, baseDir) {
   }
   const discoveredTypes = [];
   for (const filePath of files) {
-    const basename = path34.basename(filePath);
+    const basename = path35.basename(filePath);
     const typeName = basename.replace(/\.(ts|js|mts|mjs)$/, "");
     if (registry.has(typeName)) {
       continue;
@@ -13541,17 +13726,17 @@ async function discoverAssertions(registry, baseDir) {
 }
 
 // src/evaluation/registry/grader-discovery.ts
-import path35 from "node:path";
+import path36 from "node:path";
 import fg3 from "fast-glob";
 async function discoverGraders(registry, baseDir) {
   const patterns = ["*.ts", "*.js", "*.mts", "*.mjs"];
   const candidateDirs = [];
-  let dir = path35.resolve(baseDir);
-  const root = path35.parse(dir).root;
+  let dir = path36.resolve(baseDir);
+  const root = path36.parse(dir).root;
   while (dir !== root) {
-    candidateDirs.push(path35.join(dir, ".agentv", "graders"));
-    candidateDirs.push(path35.join(dir, ".agentv", "judges"));
-    dir = path35.dirname(dir);
+    candidateDirs.push(path36.join(dir, ".agentv", "graders"));
+    candidateDirs.push(path36.join(dir, ".agentv", "judges"));
+    dir = path36.dirname(dir);
   }
   let files = [];
   for (const gradersDir of candidateDirs) {
@@ -13567,7 +13752,7 @@ async function discoverGraders(registry, baseDir) {
   }
   const discoveredTypes = [];
   for (const filePath of files) {
-    const basename = path35.basename(filePath);
+    const basename = path36.basename(filePath);
     const typeName = basename.replace(/\.(ts|js|mts|mjs)$/, "");
     if (registry.has(typeName)) {
       continue;
@@ -13727,7 +13912,7 @@ function getTCritical(df) {
 // src/evaluation/workspace/file-changes.ts
 import { exec as execCallback } from "node:child_process";
 import { readdirSync as readdirSync2, statSync } from "node:fs";
-import path36 from "node:path";
+import path37 from "node:path";
 import { promisify as promisify4 } from "node:util";
 var execAsync4 = promisify4(execCallback);
 function gitExecOpts(workspacePath) {
@@ -13761,10 +13946,10 @@ async function stageNestedRepoChanges(workspacePath) {
   }
   for (const entry of entries) {
     if (entry === ".git" || entry === "node_modules") continue;
-    const childPath = path36.join(workspacePath, entry);
+    const childPath = path37.join(workspacePath, entry);
     try {
       if (!statSync(childPath).isDirectory()) continue;
-      if (!statSync(path36.join(childPath, ".git")).isDirectory()) continue;
+      if (!statSync(path37.join(childPath, ".git")).isDirectory()) continue;
     } catch {
       continue;
     }
@@ -13774,8 +13959,8 @@ async function stageNestedRepoChanges(workspacePath) {
 }
 
 // src/evaluation/workspace/manager.ts
-import { cp, mkdir as mkdir11, readdir as readdir3, rm as rm4, stat as stat5 } from "node:fs/promises";
-import path37 from "node:path";
+import { cp, mkdir as mkdir12, readdir as readdir3, rm as rm4, stat as stat5 } from "node:fs/promises";
+import path38 from "node:path";
 var TemplateNotFoundError = class extends Error {
   constructor(templatePath) {
     super(`Workspace template not found: ${templatePath}`);
@@ -13805,14 +13990,14 @@ async function isDirectory(filePath) {
 }
 function getWorkspacePath(evalRunId, caseId, workspaceRoot) {
   const root = workspaceRoot ?? getWorkspacesRoot();
-  return path37.join(root, evalRunId, caseId);
+  return path38.join(root, evalRunId, caseId);
 }
 async function copyDirectoryRecursive(src, dest) {
-  await mkdir11(dest, { recursive: true });
+  await mkdir12(dest, { recursive: true });
   const entries = await readdir3(src, { withFileTypes: true });
   for (const entry of entries) {
-    const srcPath = path37.join(src, entry.name);
-    const destPath = path37.join(dest, entry.name);
+    const srcPath = path38.join(src, entry.name);
+    const destPath = path38.join(dest, entry.name);
     if (entry.name === ".git") {
       continue;
     }
@@ -13824,7 +14009,7 @@ async function copyDirectoryRecursive(src, dest) {
   }
 }
 async function createTempWorkspace(templatePath, evalRunId, caseId, workspaceRoot) {
-  const resolvedTemplatePath = path37.resolve(templatePath);
+  const resolvedTemplatePath = path38.resolve(templatePath);
   if (!await fileExists(resolvedTemplatePath)) {
     throw new TemplateNotFoundError(resolvedTemplatePath);
   }
@@ -13873,7 +14058,7 @@ async function cleanupWorkspace(workspacePath) {
 }
 async function cleanupEvalWorkspaces(evalRunId, workspaceRoot) {
   const root = workspaceRoot ?? getWorkspacesRoot();
-  const evalDir = path37.join(root, evalRunId);
+  const evalDir = path38.join(root, evalRunId);
   if (await fileExists(evalDir)) {
     await rm4(evalDir, { recursive: true, force: true });
   }
@@ -13883,8 +14068,8 @@ async function cleanupEvalWorkspaces(evalRunId, workspaceRoot) {
 import { execFile } from "node:child_process";
 import { createHash } from "node:crypto";
 import { existsSync as existsSync2 } from "node:fs";
-import { cp as cp2, mkdir as mkdir12, readFile as readFile10, readdir as readdir4, rm as rm5, unlink, writeFile as writeFile7 } from "node:fs/promises";
-import path38 from "node:path";
+import { cp as cp2, mkdir as mkdir13, readFile as readFile10, readdir as readdir4, rm as rm5, unlink, writeFile as writeFile7 } from "node:fs/promises";
+import path39 from "node:path";
 import { promisify as promisify5 } from "node:util";
 var execFileAsync = promisify5(execFile);
 function gitEnv() {
@@ -13935,11 +14120,11 @@ function computeWorkspaceFingerprint(repos) {
   return createHash("sha256").update(JSON.stringify(canonical)).digest("hex");
 }
 async function copyDirectoryRecursive2(src, dest, skipDirs) {
-  await mkdir12(dest, { recursive: true });
+  await mkdir13(dest, { recursive: true });
   const entries = await readdir4(src, { withFileTypes: true });
   for (const entry of entries) {
-    const srcPath = path38.join(src, entry.name);
-    const destPath = path38.join(dest, entry.name);
+    const srcPath = path39.join(src, entry.name);
+    const destPath = path39.join(dest, entry.name);
     if (entry.name === ".git") {
       continue;
     }
@@ -13972,8 +14157,8 @@ var WorkspacePoolManager = class {
   async acquireWorkspace(options) {
     const { templatePath, repos, maxSlots, repoManager, poolReset } = options;
     const fingerprint = computeWorkspaceFingerprint(repos);
-    const poolDir = path38.join(this.poolRoot, fingerprint);
-    await mkdir12(poolDir, { recursive: true });
+    const poolDir = path39.join(this.poolRoot, fingerprint);
+    await mkdir13(poolDir, { recursive: true });
     const drifted = await this.checkDrift(poolDir, fingerprint);
     if (drifted) {
       console.warn(
@@ -13982,7 +14167,7 @@ var WorkspacePoolManager = class {
       await this.removeAllSlots(poolDir);
     }
     for (let i = 0; i < maxSlots; i++) {
-      const slotPath = path38.join(poolDir, `slot-${i}`);
+      const slotPath = path39.join(poolDir, `slot-${i}`);
       const lockPath = `${slotPath}.lock`;
       const locked = await this.tryLock(lockPath);
       if (!locked) {
@@ -14000,7 +14185,7 @@ var WorkspacePoolManager = class {
           poolDir
         };
       }
-      await mkdir12(slotPath, { recursive: true });
+      await mkdir13(slotPath, { recursive: true });
       if (templatePath) {
         await copyDirectoryRecursive2(templatePath, slotPath);
       }
@@ -14069,7 +14254,7 @@ var WorkspacePoolManager = class {
    * Returns false (no drift) if metadata.json doesn't exist (first use).
    */
   async checkDrift(poolDir, fingerprint) {
-    const metadataPath = path38.join(poolDir, "metadata.json");
+    const metadataPath = path39.join(poolDir, "metadata.json");
     try {
       const raw = await readFile10(metadataPath, "utf-8");
       const metadata = JSON.parse(raw);
@@ -14086,14 +14271,14 @@ var WorkspacePoolManager = class {
       repos,
       createdAt: (/* @__PURE__ */ new Date()).toISOString()
     };
-    await writeFile7(path38.join(poolDir, "metadata.json"), JSON.stringify(metadata, null, 2));
+    await writeFile7(path39.join(poolDir, "metadata.json"), JSON.stringify(metadata, null, 2));
   }
   /** Remove all slot directories and their lock files from a pool directory. */
   async removeAllSlots(poolDir) {
     const entries = await readdir4(poolDir);
     for (const entry of entries) {
       if (entry.startsWith("slot-") && !entry.endsWith(".lock")) {
-        const lockPath = path38.join(poolDir, `${entry}.lock`);
+        const lockPath = path39.join(poolDir, `${entry}.lock`);
         if (existsSync2(lockPath)) {
           try {
             const pidStr = await readFile10(lockPath, "utf-8");
@@ -14109,12 +14294,12 @@ var WorkspacePoolManager = class {
           } catch {
           }
         }
-        await rm5(path38.join(poolDir, entry), { recursive: true, force: true });
+        await rm5(path39.join(poolDir, entry), { recursive: true, force: true });
         await rm5(lockPath, { force: true }).catch(() => {
         });
       }
     }
-    await rm5(path38.join(poolDir, "metadata.json"), { force: true }).catch(() => {
+    await rm5(path39.join(poolDir, "metadata.json"), { force: true }).catch(() => {
     });
   }
   /**
@@ -14124,7 +14309,7 @@ var WorkspacePoolManager = class {
    */
   async resetSlot(slotPath, templatePath, repos, poolReset = "fast") {
     for (const repo of repos) {
-      const repoDir = path38.join(slotPath, repo.path);
+      const repoDir = path39.join(slotPath, repo.path);
       if (!existsSync2(repoDir)) {
         continue;
       }
@@ -14151,7 +14336,7 @@ var WorkspacePoolManager = class {
 // src/evaluation/workspace/repo-manager.ts
 import { execFile as execFile2 } from "node:child_process";
 import { existsSync as existsSync3 } from "node:fs";
-import path39 from "node:path";
+import path40 from "node:path";
 import { promisify as promisify6 } from "node:util";
 var execFileAsync2 = promisify6(execFile2);
 var DEFAULT_TIMEOUT_MS2 = 3e5;
@@ -14251,7 +14436,7 @@ ${lines.join("\n")}`;
    * Handles checkout, ref resolution, ancestor walking, shallow clone, sparse checkout.
    */
   async materialize(repo, workspacePath) {
-    const targetDir = path39.join(workspacePath, repo.path);
+    const targetDir = path40.join(workspacePath, repo.path);
     const sourceUrl = getSourceUrl(repo.source);
     const startedAt = Date.now();
     if (this.verbose) {
@@ -14342,7 +14527,7 @@ ${lines.join("\n")}`;
   async reset(repos, workspacePath, reset) {
     const cleanFlag = reset === "strict" ? "-fdx" : "-fd";
     for (const repo of repos) {
-      const targetDir = path39.join(workspacePath, repo.path);
+      const targetDir = path40.join(workspacePath, repo.path);
       await this.runGit(["reset", "--hard", "HEAD"], { cwd: targetDir });
       await this.runGit(["clean", cleanFlag], { cwd: targetDir });
     }
@@ -14351,16 +14536,16 @@ ${lines.join("\n")}`;
 
 // src/evaluation/workspace/resolve.ts
 import { readdir as readdir5, stat as stat6 } from "node:fs/promises";
-import path40 from "node:path";
+import path41 from "node:path";
 async function resolveWorkspaceTemplate(templatePath) {
   if (!templatePath) {
     return void 0;
   }
-  const resolved = path40.resolve(templatePath);
+  const resolved = path41.resolve(templatePath);
   const stats = await stat6(resolved);
   if (stats.isFile()) {
     return {
-      dir: path40.dirname(resolved),
+      dir: path41.dirname(resolved),
       workspaceFile: resolved
     };
   }
@@ -14372,14 +14557,14 @@ async function resolveWorkspaceTemplate(templatePath) {
   if (workspaceFiles.length === 1) {
     return {
       dir: resolved,
-      workspaceFile: path40.join(resolved, workspaceFiles[0])
+      workspaceFile: path41.join(resolved, workspaceFiles[0])
     };
   }
   if (workspaceFiles.length > 1) {
     const conventionFile = workspaceFiles.find((f) => f === "template.code-workspace");
     return {
       dir: resolved,
-      workspaceFile: conventionFile ? path40.join(resolved, conventionFile) : void 0
+      workspaceFile: conventionFile ? path41.join(resolved, conventionFile) : void 0
     };
   }
   return { dir: resolved };
@@ -14516,7 +14701,7 @@ async function runEvaluation(options) {
     );
     useCache = false;
   }
-  const evalRunId = randomUUID8();
+  const evalRunId = randomUUID9();
   const evalCases = preloadedEvalCases ?? await loadTests(evalFilePath, repoRoot, { verbose, filter });
   const filteredEvalCases = filterEvalCases(evalCases, filter);
   if (filteredEvalCases.length === 0) {
@@ -14595,7 +14780,7 @@ async function runEvaluation(options) {
   ];
   const evaluatorRegistry = buildEvaluatorRegistry(evaluators, resolveGraderProvider);
   const typeRegistry = createBuiltinRegistry();
-  const discoveryBaseDir = evalFilePath ? path41.dirname(path41.resolve(evalFilePath)) : process.cwd();
+  const discoveryBaseDir = evalFilePath ? path42.dirname(path42.resolve(evalFilePath)) : process.cwd();
   const evalDir = discoveryBaseDir;
   await discoverAssertions(typeRegistry, discoveryBaseDir);
   await discoverGraders(typeRegistry, discoveryBaseDir);
@@ -14742,7 +14927,7 @@ async function runEvaluation(options) {
     const isEmpty = dirExists ? (await readdir6(configuredStaticPath)).length === 0 : false;
     if (isYamlConfiguredPath && (!dirExists || isEmpty)) {
       if (!dirExists) {
-        await mkdir13(configuredStaticPath, { recursive: true });
+        await mkdir14(configuredStaticPath, { recursive: true });
       }
       if (workspaceTemplate) {
         await copyDirectoryRecursive(workspaceTemplate, configuredStaticPath);
@@ -14787,12 +14972,12 @@ async function runEvaluation(options) {
     }
   } else if (suiteWorkspace?.hooks || suiteWorkspace?.repos?.length && !isPerTestIsolation) {
     sharedWorkspacePath = getWorkspacePath(evalRunId, "shared");
-    await mkdir13(sharedWorkspacePath, { recursive: true });
+    await mkdir14(sharedWorkspacePath, { recursive: true });
     setupLog(`created empty shared workspace at: ${sharedWorkspacePath}`);
   }
   try {
     if (suiteWorkspaceFile && sharedWorkspacePath) {
-      const copiedWorkspaceFile = path41.join(sharedWorkspacePath, path41.basename(suiteWorkspaceFile));
+      const copiedWorkspaceFile = path42.join(sharedWorkspacePath, path42.basename(suiteWorkspaceFile));
       try {
         await stat7(copiedWorkspaceFile);
         suiteWorkspaceFile = copiedWorkspaceFile;
@@ -15374,7 +15559,7 @@ async function runEvalCase(options) {
         );
       }
       if (caseWorkspaceFile && workspacePath) {
-        const copiedFile = path41.join(workspacePath, path41.basename(caseWorkspaceFile));
+        const copiedFile = path42.join(workspacePath, path42.basename(caseWorkspaceFile));
         try {
           await stat7(copiedFile);
           caseWorkspaceFile = copiedFile;
@@ -15384,7 +15569,7 @@ async function runEvalCase(options) {
     }
     if (!workspacePath && (evalCase.workspace?.hooks || evalCase.workspace?.repos?.length) && evalRunId) {
       workspacePath = getWorkspacePath(evalRunId, evalCase.id);
-      await mkdir13(workspacePath, { recursive: true });
+      await mkdir14(workspacePath, { recursive: true });
     }
     if (evalCase.workspace?.repos?.length && workspacePath) {
       const localPathErrors = RepoManager.validateLocalPaths(evalCase.workspace.repos);
@@ -15436,10 +15621,10 @@ async function runEvalCase(options) {
       const files = evalCase.metadata.agent_skills_files;
       if (baseDir && files.length > 0) {
         for (const relPath of files) {
-          const srcPath = path41.resolve(baseDir, relPath);
-          const destPath = path41.resolve(workspacePath, relPath);
+          const srcPath = path42.resolve(baseDir, relPath);
+          const destPath = path42.resolve(workspacePath, relPath);
           try {
-            await mkdir13(path41.dirname(destPath), { recursive: true });
+            await mkdir14(path42.dirname(destPath), { recursive: true });
             await copyFile2(srcPath, destPath);
           } catch (error) {
             const message = error instanceof Error ? error.message : String(error);
@@ -16085,7 +16270,7 @@ async function runEvaluatorList(options) {
     fileChanges,
     workspacePath
   };
-  const evalFileDir = evalCase.file_paths[0] ? path41.dirname(evalCase.file_paths[0]) : process.cwd();
+  const evalFileDir = evalCase.file_paths[0] ? path42.dirname(evalCase.file_paths[0]) : process.cwd();
   const dispatchContext = {
     graderProvider,
     targetResolver,
@@ -16419,7 +16604,7 @@ function computeWeightedMean(entries) {
 
 // src/evaluation/evaluate.ts
 import { existsSync as existsSync4 } from "node:fs";
-import path42 from "node:path";
+import path43 from "node:path";
 
 // src/evaluation/providers/function-provider.ts
 function createFunctionProvider(taskFn) {
@@ -16456,7 +16641,7 @@ async function evaluate(config) {
   }
   const gitRoot = await findGitRoot(process.cwd());
   const repoRoot = gitRoot ?? process.cwd();
-  const testFilePath = config.specFile ? path42.resolve(config.specFile) : path42.join(process.cwd(), "__programmatic__.yaml");
+  const testFilePath = config.specFile ? path43.resolve(config.specFile) : path43.join(process.cwd(), "__programmatic__.yaml");
   await loadEnvHierarchy(repoRoot, testFilePath);
   let resolvedTarget;
   let taskProvider;
@@ -16577,10 +16762,10 @@ function computeSummary(results, durationMs) {
 var TARGET_FILE_CANDIDATES = [".agentv/targets.yaml", ".agentv/targets.yml"];
 async function discoverDefaultTarget(repoRoot) {
   const cwd = process.cwd();
-  const chain = buildDirectoryChain(path42.join(cwd, "_placeholder"), repoRoot);
+  const chain = buildDirectoryChain(path43.join(cwd, "_placeholder"), repoRoot);
   for (const dir of chain) {
     for (const candidate of TARGET_FILE_CANDIDATES) {
-      const targetsPath = path42.join(dir, candidate);
+      const targetsPath = path43.join(dir, candidate);
       if (!existsSync4(targetsPath)) continue;
       try {
         const definitions = await readTargetDefinitions(targetsPath);
@@ -16597,7 +16782,7 @@ async function loadEnvHierarchy(repoRoot, startPath) {
   const chain = buildDirectoryChain(startPath, repoRoot);
   const envFiles = [];
   for (const dir of chain) {
-    const envPath = path42.join(dir, ".env");
+    const envPath = path43.join(dir, ".env");
     if (existsSync4(envPath)) envFiles.push(envPath);
   }
   for (let i = 0; i < envFiles.length; i++) {
@@ -16778,8 +16963,8 @@ function buildPrompt(criteria, question, referenceAnswer) {
 }
 
 // src/evaluation/cache/response-cache.ts
-import { mkdir as mkdir14, readFile as readFile11, writeFile as writeFile8 } from "node:fs/promises";
-import path43 from "node:path";
+import { mkdir as mkdir15, readFile as readFile11, writeFile as writeFile8 } from "node:fs/promises";
+import path44 from "node:path";
 var DEFAULT_CACHE_PATH = ".agentv/cache";
 var ResponseCache = class {
   cachePath;
@@ -16797,13 +16982,13 @@ var ResponseCache = class {
   }
   async set(key, value) {
     const filePath = this.keyToPath(key);
-    const dir = path43.dirname(filePath);
-    await mkdir14(dir, { recursive: true });
+    const dir = path44.dirname(filePath);
+    await mkdir15(dir, { recursive: true });
     await writeFile8(filePath, JSON.stringify(value, null, 2), "utf8");
   }
   keyToPath(key) {
     const prefix = key.slice(0, 2);
-    return path43.join(this.cachePath, prefix, `${key}.json`);
+    return path44.join(this.cachePath, prefix, `${key}.json`);
   }
 };
 function shouldEnableCache(params) {
@@ -16937,7 +17122,7 @@ var OtelTraceExporter = class {
         );
       }
       if (this.options.traceFilePath) {
-        const { SimpleTraceFileExporter: SimpleTraceFileExporter2 } = await import("./simple-trace-file-exporter-S76DMABU.js");
+        const { SimpleTraceFileExporter: SimpleTraceFileExporter2 } = await import("./simple-trace-file-exporter-CRIO5HDZ.js");
         processors.push(
           new SimpleSpanProcessor(new SimpleTraceFileExporter2(this.options.traceFilePath))
         );
@@ -17001,6 +17186,17 @@ var OtelTraceExporter = class {
         if (result.durationMs != null)
           rootSpan.setAttribute("agentv.trace.duration_ms", result.durationMs);
         if (result.costUsd != null) rootSpan.setAttribute("agentv.trace.cost_usd", result.costUsd);
+        if (result.tokenUsage) {
+          if (result.tokenUsage.input != null) {
+            rootSpan.setAttribute("agentv.trace.token_input", result.tokenUsage.input);
+          }
+          if (result.tokenUsage.output != null) {
+            rootSpan.setAttribute("agentv.trace.token_output", result.tokenUsage.output);
+          }
+          if (result.tokenUsage.cached != null) {
+            rootSpan.setAttribute("agentv.trace.token_cached", result.tokenUsage.cached);
+          }
+        }
         if (result.trace) {
           const t = result.trace;
           rootSpan.setAttribute("agentv.trace.event_count", t.eventCount);
@@ -17103,6 +17299,7 @@ var OtelTraceExporter = class {
       tracer.startActiveSpan(
         spanName,
         { startTime: startHr },
+        parentCtx,
         (span) => {
           if (isAssistant) {
             span.setAttribute("gen_ai.operation.name", "chat");
@@ -17135,6 +17332,7 @@ var OtelTraceExporter = class {
                 tracer.startActiveSpan(
                   `execute_tool ${tc.tool}`,
                   {},
+                  msgCtx,
                   (toolSpan) => {
                     toolSpan.setAttribute("gen_ai.tool.name", tc.tool);
                     if (tc.id) toolSpan.setAttribute("gen_ai.tool.call.id", tc.id);
@@ -17175,8 +17373,12 @@ var OtelStreamingObserver = class {
   rootSpan = null;
   // biome-ignore lint/suspicious/noExplicitAny: OTel context loaded dynamically
   rootCtx = null;
+  observedChildSpans = false;
+  pendingMetrics = null;
   /** Create root eval span immediately (visible in backend right away) */
   startEvalCase(testId, target, evalSet) {
+    this.pendingMetrics = null;
+    this.observedChildSpans = false;
     const ctx = this.parentCtx ?? this.api.context.active();
     this.rootSpan = this.tracer.startSpan("agentv.eval", void 0, ctx);
     this.rootSpan.setAttribute("gen_ai.operation.name", "evaluate");
@@ -17189,8 +17391,9 @@ var OtelStreamingObserver = class {
   /** Create and immediately export a tool span */
   onToolCall(name, input, output, _durationMs, toolCallId) {
     if (!this.rootCtx) return;
+    this.observedChildSpans = true;
     this.api.context.with(this.rootCtx, () => {
-      const span = this.tracer.startSpan(`execute_tool ${name}`);
+      const span = this.tracer.startSpan(`execute_tool ${name}`, void 0, this.rootCtx);
       span.setAttribute("gen_ai.tool.name", name);
       if (toolCallId) span.setAttribute("gen_ai.tool.call.id", toolCallId);
       if (this.captureContent) {
@@ -17211,8 +17414,9 @@ var OtelStreamingObserver = class {
   /** Create and immediately export an LLM span */
   onLlmCall(model, tokenUsage) {
     if (!this.rootCtx) return;
+    this.observedChildSpans = true;
     this.api.context.with(this.rootCtx, () => {
-      const span = this.tracer.startSpan(`chat ${model}`);
+      const span = this.tracer.startSpan(`chat ${model}`, void 0, this.rootCtx);
       span.setAttribute("gen_ai.operation.name", "chat");
       span.setAttribute("gen_ai.request.model", model);
       span.setAttribute("gen_ai.response.model", model);
@@ -17227,10 +17431,53 @@ var OtelStreamingObserver = class {
       span.end();
     });
   }
+  /** Record final execution metrics before the root span is finalized. */
+  recordEvalMetrics(result) {
+    this.pendingMetrics = result;
+  }
   /** Finalize root span with score/verdict after evaluation completes */
   finalizeEvalCase(score, error) {
     if (!this.rootSpan) return;
     this.rootSpan.setAttribute("agentv.score", score);
+    if (this.pendingMetrics?.durationMs != null) {
+      this.rootSpan.setAttribute("agentv.trace.duration_ms", this.pendingMetrics.durationMs);
+    }
+    if (this.pendingMetrics?.costUsd != null) {
+      this.rootSpan.setAttribute("agentv.trace.cost_usd", this.pendingMetrics.costUsd);
+    }
+    if (this.pendingMetrics?.tokenUsage) {
+      if (this.pendingMetrics.tokenUsage.input != null) {
+        this.rootSpan.setAttribute(
+          "agentv.trace.token_input",
+          this.pendingMetrics.tokenUsage.input
+        );
+      }
+      if (this.pendingMetrics.tokenUsage.output != null) {
+        this.rootSpan.setAttribute(
+          "agentv.trace.token_output",
+          this.pendingMetrics.tokenUsage.output
+        );
+      }
+      if (this.pendingMetrics.tokenUsage.cached != null) {
+        this.rootSpan.setAttribute(
+          "agentv.trace.token_cached",
+          this.pendingMetrics.tokenUsage.cached
+        );
+      }
+    }
+    if (this.pendingMetrics?.trace) {
+      this.rootSpan.setAttribute("agentv.trace.event_count", this.pendingMetrics.trace.eventCount);
+      this.rootSpan.setAttribute(
+        "agentv.trace.tool_names",
+        Object.keys(this.pendingMetrics.trace.toolCalls).sort().join(",")
+      );
+      if (this.pendingMetrics.trace.llmCallCount != null) {
+        this.rootSpan.setAttribute(
+          "agentv.trace.llm_call_count",
+          this.pendingMetrics.trace.llmCallCount
+        );
+      }
+    }
     if (error) {
       this.rootSpan.setStatus({ code: this.api.SpanStatusCode.ERROR, message: error });
     } else {
@@ -17239,6 +17486,33 @@ var OtelStreamingObserver = class {
     this.rootSpan.end();
     this.rootSpan = null;
     this.rootCtx = null;
+    this.observedChildSpans = false;
+    this.pendingMetrics = null;
+  }
+  /** Backfill child spans from the completed result when the provider emitted no live callbacks. */
+  completeFromResult(result) {
+    this.recordEvalMetrics({
+      durationMs: result.durationMs,
+      costUsd: result.costUsd,
+      tokenUsage: result.tokenUsage,
+      trace: result.trace
+    });
+    if (this.observedChildSpans || !this.rootCtx) {
+      return;
+    }
+    const model = result.output.find((msg) => msg.role === "assistant")?.metadata?.model ?? result.target ?? "unknown";
+    this.onLlmCall(String(model), result.tokenUsage);
+    for (const message of result.output) {
+      for (const toolCall of message.toolCalls ?? []) {
+        this.onToolCall(
+          toolCall.tool,
+          toolCall.input,
+          toolCall.output,
+          toolCall.durationMs ?? 0,
+          toolCall.id
+        );
+      }
+    }
   }
   /** Return the active eval span's trace ID and span ID for Braintrust trace bridging */
   getActiveSpanIds() {