agentbox-sdk 0.1.519 → 0.1.520

package/README.md

@@ -134,6 +134,22 @@ default 30 minutes, `0` settles at the first turn end as before, `Infinity`
 waits forever. On expiry the tasks are stopped best-effort and the run
 completes with the last turn's text.
 
+### Event retention
+
+`AgentResult.events` and `AgentResult.rawEvents` replay everything a run
+emitted, which means a run holds its whole transcript — every tool result,
+every file the agent read — in memory until it settles. A host that consumes
+the live stream (`for await (const event of run)` / `run.rawEvents()`) never
+reads those arrays, so it can turn retention off:
+
+```ts
+const agent = new Agent("claude-code", { sandbox, retainEvents: false });
+```
+
+Both arrays then settle as `[]`. Streamed events, `text` and `costData` are
+unaffected — cost is accumulated as payloads arrive rather than recomputed
+from a retained transcript.
+
 ### Harness commands
 
 A run whose input starts with `/name args` carries the harness's own slash

package/dist/agents/index.d.ts

@@ -1,5 +1,5 @@
-import { o as AgentProviderName, h as AgentOptions, t as AgentRunConfig, s as AgentRun, r as AgentResult, ak as RawAgentEvent, b as AgentAttachRequest, B as AttachedRun, ar as SetupLayout, ax as UserContent } from '../types-DjHP7dFq.js';
-export { a as AgentApprovalMode, c as AgentCommandConfig, d as AgentCostData, e as AgentExecutionRequest, f as AgentLocalMcpConfig, g as AgentMcpConfig, i as AgentOptionsBase, j as AgentOptionsMap, k as AgentPermissionDecision, l as AgentPermissionKind, m as AgentPermissionResponse, n as AgentProviderAdapter, p as AgentReasoningEffort, q as AgentRemoteMcpConfig, u as AgentRunSink, v as AgentSetupRequest, w as AgentSkillConfig, x as AgentSubAgentConfig, y as AgentUserAnswer, z as AgentUserQuestion, E as ClaudeCodeAgentOptions, F as ClaudeCodeHookConfig, G as ClaudeCodeHookEvent, H as ClaudeCodeHookHandler, I as ClaudeCodeHookMatcherGroup, J as ClaudeCodeHooksConfig, K as ClaudeCodeProviderOptions, L as CodexAgentOptions, M as CodexCommandHook, N as CodexHookEvent, O as CodexHookMatcherGroup, P as CodexHooksConfig, Q as CodexModelProviderConfig, R as CodexProviderOptions, S as CodexSkillEntry, T as DataContent, U as EmbeddedSkillConfig, V as FilePart, W as HarnessCommandDescriptor, X as HarnessCommandInvocation, Y as HarnessCommandSource, _ as ImagePart, $ as MAX_HARNESS_COMMANDS, a0 as MAX_HARNESS_COMMAND_ARGUMENT_HINT, a1 as MAX_HARNESS_COMMAND_DESCRIPTION, a2 as MAX_HARNESS_COMMAND_NAME, aa as OpenCodeAgentOptions, ab as OpenCodeCommandEntry, ac as OpenCodePluginConfig, ad as OpenCodePluginEvent, ae as OpenCodePluginHookConfig, af as OpenCodeProviderOptions, ag as OpenCodeSkillEntry, ah as OpenRouterPlugin, am as RepoSkillConfig, at as TextPart, ay as UserContentPart, az as builtinHarnessCommands, aA as claudeHarnessCommands, aB as codexHarnessCommands, aC as codexSkillMention, aF as openCodeHarnessCommands, aG as parseHarnessCommandInvocation, aH as skillDirective } from '../types-DjHP7dFq.js';
+import { o as AgentProviderName, h as AgentOptions, t as AgentRunConfig, s as AgentRun, r as AgentResult, ak as RawAgentEvent, b as AgentAttachRequest, B as AttachedRun, ar as SetupLayout, ax as UserContent } from '../types-Bhntrw9V.js';
+export { a as AgentApprovalMode, c as AgentCommandConfig, d as AgentCostData, e as AgentExecutionRequest, f as AgentLocalMcpConfig, g as AgentMcpConfig, i as AgentOptionsBase, j as AgentOptionsMap, k as AgentPermissionDecision, l as AgentPermissionKind, m as AgentPermissionResponse, n as AgentProviderAdapter, p as AgentReasoningEffort, q as AgentRemoteMcpConfig, u as AgentRunSink, v as AgentSetupRequest, w as AgentSkillConfig, x as AgentSubAgentConfig, y as AgentUserAnswer, z as AgentUserQuestion, E as ClaudeCodeAgentOptions, F as ClaudeCodeHookConfig, G as ClaudeCodeHookEvent, H as ClaudeCodeHookHandler, I as ClaudeCodeHookMatcherGroup, J as ClaudeCodeHooksConfig, K as ClaudeCodeProviderOptions, L as CodexAgentOptions, M as CodexCommandHook, N as CodexHookEvent, O as CodexHookMatcherGroup, P as CodexHooksConfig, Q as CodexModelProviderConfig, R as CodexProviderOptions, S as CodexSkillEntry, T as DataContent, U as EmbeddedSkillConfig, V as FilePart, W as HarnessCommandDescriptor, X as HarnessCommandInvocation, Y as HarnessCommandSource, _ as ImagePart, $ as MAX_HARNESS_COMMANDS, a0 as MAX_HARNESS_COMMAND_ARGUMENT_HINT, a1 as MAX_HARNESS_COMMAND_DESCRIPTION, a2 as MAX_HARNESS_COMMAND_NAME, aa as OpenCodeAgentOptions, ab as OpenCodeCommandEntry, ac as OpenCodePluginConfig, ad as OpenCodePluginEvent, ae as OpenCodePluginHookConfig, af as OpenCodeProviderOptions, ag as OpenCodeSkillEntry, ah as OpenRouterPlugin, am as RepoSkillConfig, at as TextPart, ay as UserContentPart, az as builtinHarnessCommands, aA as claudeHarnessCommands, aB as codexHarnessCommands, aC as codexSkillMention, aF as openCodeHarnessCommands, aG as parseHarnessCommandInvocation, aH as skillDirective } from '../types-Bhntrw9V.js';
 import { S as Sandbox } from '../Sandbox-DcKAU-E3.js';
 export { AgentProvider } from '../enums.js';
 import 'e2b';

package/dist/agents/index.js

@@ -15,12 +15,12 @@ import {
   parseHarnessCommandInvocation,
   resolveHarnessCommand,
   skillDirective
-} from "../chunk-REVRD4IY.js";
+} from "../chunk-7PZKMK4U.js";
 import "../chunk-V4M4LPPT.js";
 import {
   AGENT_RESERVED_PORTS,
   collectAllAgentReservedPorts
-} from "../chunk-MTQ2S46C.js";
+} from "../chunk-D4LSWCFA.js";
 import "../chunk-NSJM57Z4.js";
 import {
   AgentProvider

package/dist/{chunk-REVRD4IY.js → chunk-7PZKMK4U.js}

@@ -20,7 +20,7 @@ import {
   sleep,
   time,
   waitFor
-} from "./chunk-MTQ2S46C.js";
+} from "./chunk-D4LSWCFA.js";
 import {
   shellQuote
 } from "./chunk-NSJM57Z4.js";
@@ -2034,75 +2034,83 @@ function compactCostData(costData) {
   };
   return Object.keys(compacted).length > 0 ? compacted : null;
 }
-function extractClaudeCostData(events) {
-  for (let i = events.length - 1; i >= 0; i--) {
-    const event = events[i];
-    if (!event) {
-      continue;
-    }
-    if (event.type !== "result") {
-      continue;
-    }
-    const totalCost = typeof event.total_cost_usd === "number" ? event.total_cost_usd : void 0;
-    const usage = {};
-    const modelUsage = asRecord(event.modelUsage);
-    if (modelUsage) {
-      for (const value of Object.values(modelUsage)) {
-        mergeUsage(usage, asRecord(value));
+function createClaudeCostAccumulator() {
+  let lastResult;
+  return {
+    add(event) {
+      if (event && event.type === "result") lastResult = event;
+    },
+    result() {
+      const event = lastResult;
+      if (!event) return null;
+      const totalCost = typeof event.total_cost_usd === "number" ? event.total_cost_usd : void 0;
+      const usage = {};
+      const modelUsage = asRecord(event.modelUsage);
+      if (modelUsage) {
+        for (const value of Object.values(modelUsage)) {
+          mergeUsage(usage, asRecord(value));
+        }
+      } else {
+        mergeUsage(usage, asRecord(event.usage));
       }
-    } else {
-      mergeUsage(usage, asRecord(event.usage));
-    }
-    return compactCostData({
-      ...totalCost !== void 0 ? { total_cost_usd: totalCost } : {},
-      duration_ms: typeof event.duration_ms === "number" ? event.duration_ms : void 0,
-      duration_api_ms: typeof event.duration_api_ms === "number" ? event.duration_api_ms : void 0,
-      num_turns: typeof event.num_turns === "number" ? event.num_turns : void 0,
-      usage
-    });
-  }
-  return null;
+      return compactCostData({
+        ...totalCost !== void 0 ? { total_cost_usd: totalCost } : {},
+        duration_ms: typeof event.duration_ms === "number" ? event.duration_ms : void 0,
+        duration_api_ms: typeof event.duration_api_ms === "number" ? event.duration_api_ms : void 0,
+        num_turns: typeof event.num_turns === "number" ? event.num_turns : void 0,
+        usage
+      });
+    }
+  };
 }
-function extractCodexCostData(events) {
+function createCodexCostAccumulator() {
   const usage = {};
   let sawUsage = false;
-  for (const event of events) {
-    const params = asRecord(event.params);
-    const tokenUsage = asRecord(params?.tokenUsage);
-    const turn = asRecord(params?.turn);
-    const turnTokenUsage = asRecord(turn?.tokenUsage);
-    const usageCandidate = asRecord(event.usage) ?? asRecord(params?.usage) ?? asRecord(tokenUsage?.last) ?? asRecord(tokenUsage?.total) ?? tokenUsage ?? asRecord(turn?.usage) ?? asRecord(turnTokenUsage?.last) ?? asRecord(turnTokenUsage?.total) ?? turnTokenUsage;
-    if (usageCandidate) {
-      sawUsage = true;
-      mergeUsage(usage, usageCandidate);
+  return {
+    add(event) {
+      if (!event) return;
+      const params = asRecord(event.params);
+      const tokenUsage = asRecord(params?.tokenUsage);
+      const turn = asRecord(params?.turn);
+      const turnTokenUsage = asRecord(turn?.tokenUsage);
+      const usageCandidate = asRecord(event.usage) ?? asRecord(params?.usage) ?? asRecord(tokenUsage?.last) ?? asRecord(tokenUsage?.total) ?? tokenUsage ?? asRecord(turn?.usage) ?? asRecord(turnTokenUsage?.last) ?? asRecord(turnTokenUsage?.total) ?? turnTokenUsage;
+      if (usageCandidate) {
+        sawUsage = true;
+        mergeUsage(usage, usageCandidate);
+      }
+    },
+    result() {
+      return sawUsage ? compactCostData({ usage }) : null;
     }
-  }
-  return sawUsage ? compactCostData({ usage }) : null;
+  };
 }
-function extractOpenCodeCostData(events) {
+function createOpenCodeCostAccumulator() {
   const usage = {};
   let totalCost = 0;
   let sawCostData = false;
-  for (const event of events) {
-    const properties = asRecord(event.properties);
-    const part = asRecord(properties?.part) ?? asRecord(event.part);
-    if (part?.type !== "step-finish") {
-      continue;
-    }
-    if (typeof part.cost === "number") {
-      totalCost += part.cost;
-      sawCostData = true;
-    }
-    const tokens = asRecord(part.tokens);
-    if (tokens) {
-      sawCostData = true;
-      mergeUsage(usage, tokens);
+  return {
+    add(event) {
+      if (!event) return;
+      const properties = asRecord(event.properties);
+      const part = asRecord(properties?.part) ?? asRecord(event.part);
+      if (part?.type !== "step-finish") return;
+      if (typeof part.cost === "number") {
+        totalCost += part.cost;
+        sawCostData = true;
+      }
+      const tokens = asRecord(part.tokens);
+      if (tokens) {
+        sawCostData = true;
+        mergeUsage(usage, tokens);
+      }
+    },
+    result() {
+      return sawCostData ? compactCostData({
+        ...totalCost > 0 ? { total_cost_usd: totalCost } : {},
+        usage
+      }) : null;
     }
-  }
-  return sawCostData ? compactCostData({
-    ...totalCost > 0 ? { total_cost_usd: totalCost } : {},
-    usage
-  }) : null;
+  };
 }
 
 // src/agents/background-tasks.ts
@@ -2362,7 +2370,7 @@ var BackgroundWaitFinish = class {
 };
 
 // src/agents/providers/claude-code.ts
-var DAEMON_PROTOCOL_VERSION = "10";
+var DAEMON_PROTOCOL_VERSION = "11";
 var DAEMON_PORT = 43180;
 var DAEMON_PATH = "/tmp/agentbox/claude-code/daemon.mjs";
 var DAEMON_LOG_PATH = "/tmp/agentbox/claude-code/daemon.log";
@@ -3037,7 +3045,7 @@ async function handleAbort(_req, res, runId) {
     res.end();
     return;
   }
-  await run.query.interrupt().catch(() => {});
+  await run.query?.interrupt().catch(() => {});
   res.writeHead(204);
   res.end();
 }
@@ -3053,6 +3061,20 @@ async function handleDelete(_req, res, runId) {
   res.end();
 }
 
+// Every route runs through this. A handler that throws must never become an
+// unhandled rejection: this process owns every live run in the sandbox plus
+// whatever background work is parked here, and a default-exit would take all
+// of it down over one bad request.
+function guard(promise, res) {
+  promise.catch((error) => {
+    console.error("[claude-code-daemon] request failed:", error?.stack ?? error);
+    try {
+      if (!res.headersSent) res.writeHead(500, { "content-type": "text/plain" });
+      if (!res.writableEnded) res.end();
+    } catch {}
+  });
+}
+
 const server = http.createServer((req, res) => {
   if (req.method === "GET" && req.url === "/__version") {
     res.writeHead(200, { "content-type": "text/plain" });
@@ -3070,42 +3092,58 @@ const server = http.createServer((req, res) => {
   const url = req.url ?? "";
   const permissionRoute = url.match(/^\\/runs\\/([^/]+)\\/permission$/);
   if (req.method === "POST" && permissionRoute) {
-    handlePermission(req, res, decodeURIComponent(permissionRoute[1])).catch(() => { if (!res.headersSent) res.writeHead(400); res.end(); });
+    guard(handlePermission(req, res, decodeURIComponent(permissionRoute[1])), res);
     return;
   }
   let m;
   if (req.method === "POST" && (m = url.match(/^\\/runs\\/([^/]+)\\/start$/))) {
-    handleStart(req, res, decodeURIComponent(m[1]));
+    guard(handleStart(req, res, decodeURIComponent(m[1])), res);
     return;
   }
   if (req.method === "POST" && (m = url.match(/^\\/runs\\/([^/]+)\\/sendMessage$/))) {
-    handleSendMessage(req, res, decodeURIComponent(m[1]));
+    guard(handleSendMessage(req, res, decodeURIComponent(m[1])), res);
     return;
   }
   if (req.method === "POST" && (m = url.match(/^\\/runs\\/([^/]+)\\/abort$/))) {
-    handleAbort(req, res, decodeURIComponent(m[1]));
+    guard(handleAbort(req, res, decodeURIComponent(m[1])), res);
     return;
   }
   if (req.method === "POST" && (m = url.match(/^\\/runs\\/([^/]+)\\/park$/))) {
-    handlePark(req, res, decodeURIComponent(m[1])).catch(() => { if (!res.headersSent) res.writeHead(400); res.end(); });
+    guard(handlePark(req, res, decodeURIComponent(m[1])), res);
     return;
   }
   if (req.method === "DELETE" && (m = url.match(/^\\/runs\\/([^/]+)$/))) {
-    handleDelete(req, res, decodeURIComponent(m[1]));
+    guard(handleDelete(req, res, decodeURIComponent(m[1])), res);
     return;
   }
   res.writeHead(404, { "content-type": "text/plain" });
   res.end("not found");
 });
 
+server.on("error", (error) => {
+  console.error("[claude-code-daemon] server error:", error?.stack ?? error);
+});
+
 server.listen(port, "0.0.0.0", () => {
   console.error("[claude-code-daemon] listening on :" + port + " v" + VERSION);
 });
 
+// Last line of defence. Node's default is to exit on either of these, which
+// would tear down every live run and every parked CLI in this sandbox over
+// one stray error (a write to a socket that just died, a malformed frame).
+// Staying up costs nothing: a run whose own stream is broken still unwinds
+// through its sink's "close" handler.
+process.on("unhandledRejection", (reason) => {
+  console.error("[claude-code-daemon] unhandled rejection:", reason?.stack ?? reason);
+});
+process.on("uncaughtException", (error) => {
+  console.error("[claude-code-daemon] uncaught exception:", error?.stack ?? error);
+});
+
 const shutdown = () => {
   for (const r of liveRuns.values()) {
     r.prompt.end();
-    r.query.interrupt().catch(() => {});
+    r.query?.interrupt().catch(() => {});
   }
   server.close();
   setTimeout(() => process.exit(0), 100).unref();
@@ -3396,7 +3434,10 @@ var ClaudeCodeAgentAdapter = class {
     const executeStartedAt = Date.now();
     debugClaude("execute() start runId=%s", request.runId);
     const sandbox = request.options.sandbox;
-    if (!sandbox) return executeNativeClaude(request, sink);
+    if (!sandbox) {
+      await executeNativeClaude(request, sink);
+      return;
+    }
     const claudeDir = claudeConfigDir(request.options);
     const settingsPath = path8.join(claudeDir, "settings.json");
     const mcpConfigPath = path8.join(claudeDir, "agentbox-mcp.json");
@@ -3538,55 +3579,71 @@ ${serialized}` : serialized;
       )
     );
     const tracker = new BackgroundTaskTracker();
-    const permissionMessages = async function* () {
-      for await (const item of parseNdjsonStream(response.body)) {
-        const control = item;
-        if (!control._permission) {
-          yield item;
-          continue;
+    const pendingReplies = /* @__PURE__ */ new Set();
+    let replyError;
+    const answerPermission = async (ask) => {
+      const isQuestion = ask.toolName === "AskUserQuestion";
+      const isPlan = ask.toolName === "ExitPlanMode";
+      const answer = await sink.requestPermission({
+        type: "permission.requested",
+        provider: request.provider,
+        runId: request.runId,
+        timestamp: (/* @__PURE__ */ new Date()).toISOString(),
+        requestId: ask.requestId,
+        kind: isQuestion ? "question" : isPlan ? "plan" : "tool",
+        toolName: ask.toolName,
+        title: isQuestion ? "Your input is needed" : isPlan ? "Review the plan" : ask.title ?? `Allow ${ask.toolName}?`,
+        input: ask.input,
+        ...isQuestion ? { questions: normalizeUserQuestions("claude-code", ask.input) } : {}
+      });
+      const reply = await fetch(
+        `${baseUrl}/runs/${encodeURIComponent(request.runId)}/permission`,
+        {
+          method: "POST",
+          headers: { "content-type": "application/json", ...authHeaders },
+          signal: fetchAbort.signal,
+          body: JSON.stringify({
+            requestId: ask.requestId,
+            response: answer.decision === "allow" ? {
+              behavior: "allow",
+              updatedInput: isQuestion ? {
+                ...ask.input,
+                answers: questionReply(
+                  "claude-code",
+                  ask.input,
+                  answer.answers ?? []
+                )
+              } : ask.input
+            } : {
+              behavior: "deny",
+              message: "The user declined this request."
+            }
+          })
         }
-        const ask = control._permission;
-        const isQuestion = ask.toolName === "AskUserQuestion";
-        const isPlan = ask.toolName === "ExitPlanMode";
-        const answer = await sink.requestPermission({
-          type: "permission.requested",
-          provider: request.provider,
-          runId: request.runId,
-          timestamp: (/* @__PURE__ */ new Date()).toISOString(),
-          requestId: ask.requestId,
-          kind: isQuestion ? "question" : isPlan ? "plan" : "tool",
-          toolName: ask.toolName,
-          title: isQuestion ? "Your input is needed" : isPlan ? "Review the plan" : ask.title ?? `Allow ${ask.toolName}?`,
-          input: ask.input,
-          ...isQuestion ? { questions: normalizeUserQuestions("claude-code", ask.input) } : {}
-        });
-        const reply = await fetch(
-          `${baseUrl}/runs/${encodeURIComponent(request.runId)}/permission`,
-          {
-            method: "POST",
-            headers: { "content-type": "application/json", ...authHeaders },
-            signal: fetchAbort.signal,
-            body: JSON.stringify({
-              requestId: ask.requestId,
-              response: answer.decision === "allow" ? {
-                behavior: "allow",
-                updatedInput: isQuestion ? {
-                  ...ask.input,
-                  answers: questionReply(
-                    "claude-code",
-                    ask.input,
-                    answer.answers ?? []
-                  )
-                } : ask.input
-              } : {
-                behavior: "deny",
-                message: "The user declined this request."
-              }
-            })
+      );
+      if (!reply.ok)
+        throw new Error(`Claude permission response failed: ${reply.status}`);
+    };
+    const permissionMessages = async function* () {
+      try {
+        for await (const item of parseNdjsonStream(response.body)) {
+          if (replyError !== void 0) throw replyError;
+          const control = item;
+          if (!control._permission) {
+            yield item;
+            continue;
           }
-        );
-        if (!reply.ok)
-          throw new Error(`Claude permission response failed: ${reply.status}`);
+          const pending = answerPermission(control._permission).catch(
+            (error) => {
+              if (replyError === void 0 && !cancelled) replyError = error;
+            }
+          );
+          pendingReplies.add(pending);
+          void pending.finally(() => pendingReplies.delete(pending));
+        }
+        if (replyError !== void 0) throw replyError;
+      } finally {
+        await Promise.allSettled([...pendingReplies]);
       }
     };
     const parking = request.options.provider?.parkBackgroundWork;
@@ -3626,7 +3683,6 @@ ${serialized}` : serialized;
       },
       tracker
     );
-    return async () => void 0;
   }
   /**
    * Stateless abort. POSTs to the in-sandbox daemon's
@@ -3695,7 +3751,7 @@ async function consumeClaudeMessages(request, sink, messages, executeStartedAt,
   let firstTextDeltaLogged = false;
   let lastTerminalReason;
   let lastIsError = false;
-  const rawPayloads = [];
+  const cost = createClaudeCostAccumulator();
   const timeoutMs = resolveBackgroundTaskTimeoutMs(
     request.options.backgroundTaskTimeoutMs
   );
@@ -3709,6 +3765,7 @@ async function consumeClaudeMessages(request, sink, messages, executeStartedAt,
   let leave = cleanup;
   let foreignResults = 0;
   let nothingParked = false;
+  const queuedCommands = /* @__PURE__ */ new Set();
   let lastTasksKey = JSON.stringify({
     tasks: [],
     waiting: false,
@@ -3817,7 +3874,7 @@ async function consumeClaudeMessages(request, sink, messages, executeStartedAt,
         }
       }
       const message = item;
-      rawPayloads.push(message);
+      cost.add(message);
       sink.emitRaw(toRawEvent(request.runId, message, message.type));
       if (tracker.ingest(message) && pendingWait) {
         debugClaude(
@@ -3834,6 +3891,15 @@ async function consumeClaudeMessages(request, sink, messages, executeStartedAt,
       );
       if (sessionState === "idle" && pendingWait && (tracker.liveTasks().length === 0 || await tryPark()))
         break;
+      if (message.type === "command_lifecycle") {
+        const lifecycle = message;
+        const id = lifecycle.command_uuid;
+        if (id) {
+          if (lifecycle.state === "queued") queuedCommands.add(id);
+          else queuedCommands.delete(id);
+        }
+        continue;
+      }
       if (message.type === "system") {
         const sub = message.subtype;
         if (sub === "init") {
@@ -3959,6 +4025,16 @@ async function consumeClaudeMessages(request, sink, messages, executeStartedAt,
           streamedThinkingChars = 0;
           continue;
         }
+        const origin = message.origin;
+        if (origin?.kind === "task-notification" && queuedCommands.size > 0) {
+          debugClaude(
+            "\u2605 notification turn ended while this run's prompt is still queued; not its answer"
+          );
+          sawResult = true;
+          accumulatedText = "";
+          streamedThinkingChars = 0;
+          continue;
+        }
         sawResult = true;
         const result = message;
         lastTerminalReason = result.terminal_reason;
@@ -4028,7 +4104,7 @@ async function consumeClaudeMessages(request, sink, messages, executeStartedAt,
       );
       sink.cancel({
         text: finalText,
-        costData: extractClaudeCostData(rawPayloads)
+        costData: cost.result()
       });
     } else if (isError) {
       debugClaude(
@@ -4056,7 +4132,7 @@ async function consumeClaudeMessages(request, sink, messages, executeStartedAt,
       );
       sink.complete({
         text: finalText,
-        costData: extractClaudeCostData(rawPayloads),
+        costData: cost.result(),
         // Told, not guessed: an attach that found nothing parked ran no turn
         // at all, which from the outside looks exactly like a real tool-only
         // turn that produced no answer text and no usage.
@@ -4278,7 +4354,6 @@ async function executeNativeClaude(request, sink, wait = {}) {
   } finally {
     await stop();
   }
-  return stop;
 }
 
 // src/agents/providers/codex.ts
@@ -5570,7 +5645,7 @@ var CodexAgentAdapter = class {
       };
     };
     sink.onMessage(sendTurn);
-    const rawPayloads = [];
+    const cost = createCodexCostAccumulator();
     const pendingFileChanges = /* @__PURE__ */ new Map();
     const fileItemKey = (params, itemId) => typeof params?.threadId === "string" && typeof params.turnId === "string" && typeof itemId === "string" ? `${params.threadId}:${params.turnId}:${itemId}` : void 0;
     let streamedText = "";
@@ -5642,7 +5717,7 @@ var CodexAgentAdapter = class {
             );
           }
           const raw = toRawEvent2(request.runId, message, message.method);
-          rawPayloads.push(message);
+          cost.add(message);
           sink.emitRaw(raw);
           const item = message.params?.item;
           const itemKey = fileItemKey(message.params, item?.id);
@@ -5808,7 +5883,7 @@ var CodexAgentAdapter = class {
         client.bindThread(threadResponse.thread.id);
       }
       sink.setSessionId(threadResponse.thread.id);
-      rawPayloads.push(threadResponse);
+      cost.add(threadResponse);
       sink.emitRaw(
         toRawEvent2(request.runId, threadResponse, threadResultEventName)
       );
@@ -5870,7 +5945,7 @@ var CodexAgentAdapter = class {
           );
           sink.cancel({
             text: streamedText || lastTurn?.text || void 0,
-            costData: extractCodexCostData(rawPayloads)
+            costData: cost.result()
           });
         } else {
           sink.fail(completionError);
@@ -5883,21 +5958,20 @@ var CodexAgentAdapter = class {
             Date.now() - executeStartedAt,
             interrupted
           );
-          sink.cancel({ text: text2, costData: extractCodexCostData(rawPayloads) });
+          sink.cancel({ text: text2, costData: cost.result() });
         } else {
           debugCodex(
             "\u2605 run.completed (%dms since execute start) chars=%d",
             Date.now() - executeStartedAt,
             text2?.length ?? 0
           );
-          sink.complete({ costData: extractCodexCostData(rawPayloads) });
+          sink.complete({ costData: cost.result() });
         }
       }
     } finally {
       pendingWait?.clear();
       await runtime.cleanup().catch(() => void 0);
     }
-    return async () => void 0;
   }
   /**
    * Stateless abort. Calls `turn/interrupt` against the in-sandbox
@@ -6650,7 +6724,7 @@ var OpenCodeAgentAdapter = class {
         runId: request.runId
       })
     );
-    const rawPayloads = [];
+    const cost = createOpenCodeCostAccumulator();
     const sseAbort = new AbortController();
     let sseTask;
     let capturedSessionId;
@@ -6901,7 +6975,7 @@ var OpenCodeAgentAdapter = class {
               `sse:${event.event ?? "message"}`
             );
             if (payload && typeof payload === "object" && !Array.isArray(payload)) {
-              rawPayloads.push(payload);
+              cost.add(payload);
             }
             sink.emitRaw(raw);
             const eventType = typeof payload?.type === "string" ? String(payload.type) : event.event;
@@ -7146,7 +7220,7 @@ var OpenCodeAgentAdapter = class {
         )
       );
       if (createdSession) {
-        rawPayloads.push(createdSession);
+        cost.add(createdSession);
       }
       sink.emitEvent(
         createNormalizedEvent("message.started", {
@@ -7325,7 +7399,7 @@ var OpenCodeAgentAdapter = class {
         );
         sink.cancel({
           text: streamedTextFromSse || void 0,
-          costData: extractOpenCodeCostData(rawPayloads)
+          costData: cost.result()
         });
       } else if (sessionErrorFromSse) {
         sink.fail(sessionErrorFromSse);
@@ -7366,7 +7440,7 @@ var OpenCodeAgentAdapter = class {
         );
         sink.complete({
           text: lastAssistantText,
-          costData: extractOpenCodeCostData(rawPayloads)
+          costData: cost.result()
         });
       } else if (sseSilent) {
         sink.fail(
@@ -7382,7 +7456,6 @@ var OpenCodeAgentAdapter = class {
         await sseTask.catch(() => void 0);
       }
     }
-    return async () => void 0;
   }
   /**
    * Stateless abort. Resolve the in-sandbox base URL via
@@ -7536,6 +7609,12 @@ function prepareAgentOptions(_provider, options) {
 var AgentRunController = class {
   id;
   provider;
+  /**
+   * Whether to keep every event for {@link AgentResult}. Off, the queues
+   * still deliver to live consumers but nothing is retained — see
+   * {@link AgentOptionsBase.retainEvents}.
+   */
+  retainEvents;
   sessionId;
   raw;
   sessionIdReady;
@@ -7557,9 +7636,10 @@ var AgentRunController = class {
   resolveSessionIdReady;
   rejectSessionIdReady;
   resolveFinished;
-  constructor(provider, id) {
+  constructor(provider, id, retainEvents = true) {
     this.provider = provider;
     this.id = id;
+    this.retainEvents = retainEvents;
     let resolveFinished;
     let resolveSessionIdReady;
     let rejectSessionIdReady;
@@ -7598,11 +7678,11 @@ var AgentRunController = class {
     this.resolveSessionIdReady(sessionId);
   }
   emitRaw(event) {
-    this.rawEventsList.push(event);
+    if (this.retainEvents) this.rawEventsList.push(event);
     this.rawQueue.push(event);
   }
   pushEvent(event) {
-    this.events.push(event);
+    if (this.retainEvents) this.events.push(event);
     if (event.type === "text.delta") {
       this.text += event.delta;
     } else if ((event.type === "message.completed" || event.type === "run.completed") && event.text) {
@@ -7958,7 +8038,11 @@ var Agent = class {
     const runId = runConfig.runId ?? randomUUID3();
     const streamCalledAt = Date.now();
     debugAgent("stream() provider=%s runId=%s", this.provider, runId);
-    const run = new AgentRunController(this.provider, runId);
+    const run = new AgentRunController(
+      this.provider,
+      runId,
+      this.options.retainEvents !== false
+    );
     const setupPromise = this.setupPromise;
     const provider = this.provider;
     const options = this.options;
@@ -7974,15 +8058,12 @@ var Agent = class {
           options,
           run: buildRunConfig(options, runConfig)
         };
-        const cleanup = await adapter.execute(request, run);
+        await adapter.execute(request, run);
         debugAgent(
           "adapter.execute() returned for runId=%s after %dms",
           runId,
           Date.now() - streamCalledAt
         );
-        run.setAbort(async () => {
-          await cleanup();
-        });
       } catch (error) {
         run.fail(error);
       }

package/dist/{chunk-MTQ2S46C.js → chunk-D4LSWCFA.js}

@@ -75,8 +75,18 @@ var AsyncQueue = class {
       this.resolvers.shift()?.resolve({ done: true, value: void 0 });
     }
   }
+  /**
+   * Fail the queue for whatever has not been delivered yet.
+   *
+   * A queue that already {@link finish}ed stays finished: every event reached
+   * the consumer and the transport merely closed afterwards (a websocket
+   * "error" landing right behind its "close"), which is not a run failure.
+   * Buffered items also survive — {@link next} drains them before it throws —
+   * so a transport that dies mid-stream still hands over everything it
+   * received instead of discarding the tail.
+   */
   fail(error) {
-    if (this.failure) {
+    if (this.closed || this.failure) {
       return;
     }
     this.failure = asError(error);
@@ -85,13 +95,13 @@ var AsyncQueue = class {
     }
   }
   async next() {
-    if (this.failure) {
-      throw this.failure;
-    }
     if (this.items.length > 0) {
       const value = this.items.shift();
       return { done: false, value };
     }
+    if (this.failure) {
+      throw this.failure;
+    }
     if (this.closed) {
       return { done: true, value: void 0 };
     }

package/dist/{chunk-76AFK7XN.js → chunk-ZQUE5KT6.js}

@@ -10,7 +10,7 @@ import {
   sleep,
   suppressUnhandledRejection,
   time
-} from "./chunk-MTQ2S46C.js";
+} from "./chunk-D4LSWCFA.js";
 import {
   shellQuote,
   toShellCommand

package/dist/events/index.d.ts

@@ -1,4 +1,4 @@
-export { A as AISDKEvent, C as BackgroundTask, D as BackgroundTasksEvent, Z as HarnessCommandsEvent, a3 as MessageCompletedEvent, a4 as MessageInjectedEvent, a5 as MessageStartedEvent, a6 as NativePlanEvent, a7 as NormalizedAgentEvent, a8 as NormalizedAgentEventBase, a9 as NormalizedAgentEventType, ai as PermissionRequestedEvent, aj as PermissionResolvedEvent, ak as RawAgentEvent, al as ReasoningDeltaEvent, an as RunCancelledEvent, ao as RunCompletedEvent, ap as RunErrorEvent, aq as RunStartedEvent, as as TextDeltaEvent, au as ToolCallCompletedEvent, av as ToolCallDeltaEvent, aw as ToolCallStartedEvent, aD as createNormalizedEvent, aE as normalizeRawAgentEvent, aI as toAISDKEvent, aJ as toAISDKStream } from '../types-DjHP7dFq.js';
+export { A as AISDKEvent, C as BackgroundTask, D as BackgroundTasksEvent, Z as HarnessCommandsEvent, a3 as MessageCompletedEvent, a4 as MessageInjectedEvent, a5 as MessageStartedEvent, a6 as NativePlanEvent, a7 as NormalizedAgentEvent, a8 as NormalizedAgentEventBase, a9 as NormalizedAgentEventType, ai as PermissionRequestedEvent, aj as PermissionResolvedEvent, ak as RawAgentEvent, al as ReasoningDeltaEvent, an as RunCancelledEvent, ao as RunCompletedEvent, ap as RunErrorEvent, aq as RunStartedEvent, as as TextDeltaEvent, au as ToolCallCompletedEvent, av as ToolCallDeltaEvent, aw as ToolCallStartedEvent, aD as createNormalizedEvent, aE as normalizeRawAgentEvent, aI as toAISDKEvent, aJ as toAISDKStream } from '../types-Bhntrw9V.js';
 import { AgentProvider } from '../enums.js';
 import '../Sandbox-DcKAU-E3.js';
 import 'e2b';

package/dist/index.d.ts

@@ -1,4 +1,4 @@
-export { A as AISDKEvent, a as AgentApprovalMode, b as AgentAttachRequest, c as AgentCommandConfig, d as AgentCostData, e as AgentExecutionRequest, f as AgentLocalMcpConfig, g as AgentMcpConfig, h as AgentOptions, i as AgentOptionsBase, j as AgentOptionsMap, k as AgentPermissionDecision, l as AgentPermissionKind, m as AgentPermissionResponse, n as AgentProviderAdapter, o as AgentProviderName, p as AgentReasoningEffort, q as AgentRemoteMcpConfig, r as AgentResult, s as AgentRun, t as AgentRunConfig, u as AgentRunSink, v as AgentSetupRequest, w as AgentSkillConfig, x as AgentSubAgentConfig, y as AgentUserAnswer, z as AgentUserQuestion, B as AttachedRun, C as BackgroundTask, D as BackgroundTasksEvent, E as ClaudeCodeAgentOptions, F as ClaudeCodeHookConfig, G as ClaudeCodeHookEvent, H as ClaudeCodeHookHandler, I as ClaudeCodeHookMatcherGroup, J as ClaudeCodeHooksConfig, K as ClaudeCodeProviderOptions, L as CodexAgentOptions, M as CodexCommandHook, N as CodexHookEvent, O as CodexHookMatcherGroup, P as CodexHooksConfig, Q as CodexModelProviderConfig, R as CodexProviderOptions, S as CodexSkillEntry, T as DataContent, U as EmbeddedSkillConfig, V as FilePart, W as HarnessCommandDescriptor, X as HarnessCommandInvocation, Y as HarnessCommandSource, Z as HarnessCommandsEvent, _ as ImagePart, $ as MAX_HARNESS_COMMANDS, a0 as MAX_HARNESS_COMMAND_ARGUMENT_HINT, a1 as MAX_HARNESS_COMMAND_DESCRIPTION, a2 as MAX_HARNESS_COMMAND_NAME, a3 as MessageCompletedEvent, a4 as MessageInjectedEvent, a5 as MessageStartedEvent, a6 as NativePlanEvent, a7 as NormalizedAgentEvent, a8 as NormalizedAgentEventBase, a9 as NormalizedAgentEventType, aa as OpenCodeAgentOptions, ab as OpenCodeCommandEntry, ac as OpenCodePluginConfig, ad as OpenCodePluginEvent, ae as OpenCodePluginHookConfig, af as OpenCodeProviderOptions, ag as OpenCodeSkillEntry, ah as OpenRouterPlugin, ai as PermissionRequestedEvent, aj as PermissionResolvedEvent, ak as RawAgentEvent, al as ReasoningDeltaEvent, am as RepoSkillConfig, an as RunCancelledEvent, ao as RunCompletedEvent, ap as RunErrorEvent, aq as RunStartedEvent, ar as SetupLayout, as as TextDeltaEvent, at as TextPart, au as ToolCallCompletedEvent, av as ToolCallDeltaEvent, aw as ToolCallStartedEvent, ax as UserContent, ay as UserContentPart, az as builtinHarnessCommands, aA as claudeHarnessCommands, aB as codexHarnessCommands, aC as codexSkillMention, aD as createNormalizedEvent, aE as normalizeRawAgentEvent, aF as openCodeHarnessCommands, aG as parseHarnessCommandInvocation, aH as skillDirective, aI as toAISDKEvent, aJ as toAISDKStream } from './types-DjHP7dFq.js';
+export { A as AISDKEvent, a as AgentApprovalMode, b as AgentAttachRequest, c as AgentCommandConfig, d as AgentCostData, e as AgentExecutionRequest, f as AgentLocalMcpConfig, g as AgentMcpConfig, h as AgentOptions, i as AgentOptionsBase, j as AgentOptionsMap, k as AgentPermissionDecision, l as AgentPermissionKind, m as AgentPermissionResponse, n as AgentProviderAdapter, o as AgentProviderName, p as AgentReasoningEffort, q as AgentRemoteMcpConfig, r as AgentResult, s as AgentRun, t as AgentRunConfig, u as AgentRunSink, v as AgentSetupRequest, w as AgentSkillConfig, x as AgentSubAgentConfig, y as AgentUserAnswer, z as AgentUserQuestion, B as AttachedRun, C as BackgroundTask, D as BackgroundTasksEvent, E as ClaudeCodeAgentOptions, F as ClaudeCodeHookConfig, G as ClaudeCodeHookEvent, H as ClaudeCodeHookHandler, I as ClaudeCodeHookMatcherGroup, J as ClaudeCodeHooksConfig, K as ClaudeCodeProviderOptions, L as CodexAgentOptions, M as CodexCommandHook, N as CodexHookEvent, O as CodexHookMatcherGroup, P as CodexHooksConfig, Q as CodexModelProviderConfig, R as CodexProviderOptions, S as CodexSkillEntry, T as DataContent, U as EmbeddedSkillConfig, V as FilePart, W as HarnessCommandDescriptor, X as HarnessCommandInvocation, Y as HarnessCommandSource, Z as HarnessCommandsEvent, _ as ImagePart, $ as MAX_HARNESS_COMMANDS, a0 as MAX_HARNESS_COMMAND_ARGUMENT_HINT, a1 as MAX_HARNESS_COMMAND_DESCRIPTION, a2 as MAX_HARNESS_COMMAND_NAME, a3 as MessageCompletedEvent, a4 as MessageInjectedEvent, a5 as MessageStartedEvent, a6 as NativePlanEvent, a7 as NormalizedAgentEvent, a8 as NormalizedAgentEventBase, a9 as NormalizedAgentEventType, aa as OpenCodeAgentOptions, ab as OpenCodeCommandEntry, ac as OpenCodePluginConfig, ad as OpenCodePluginEvent, ae as OpenCodePluginHookConfig, af as OpenCodeProviderOptions, ag as OpenCodeSkillEntry, ah as OpenRouterPlugin, ai as PermissionRequestedEvent, aj as PermissionResolvedEvent, ak as RawAgentEvent, al as ReasoningDeltaEvent, am as RepoSkillConfig, an as RunCancelledEvent, ao as RunCompletedEvent, ap as RunErrorEvent, aq as RunStartedEvent, ar as SetupLayout, as as TextDeltaEvent, at as TextPart, au as ToolCallCompletedEvent, av as ToolCallDeltaEvent, aw as ToolCallStartedEvent, ax as UserContent, ay as UserContentPart, az as builtinHarnessCommands, aA as claudeHarnessCommands, aB as codexHarnessCommands, aC as codexSkillMention, aD as createNormalizedEvent, aE as normalizeRawAgentEvent, aF as openCodeHarnessCommands, aG as parseHarnessCommandInvocation, aH as skillDirective, aI as toAISDKEvent, aJ as toAISDKStream } from './types-Bhntrw9V.js';
 export { AGENT_RESERVED_PORTS, Agent, HarnessCapabilities, HarnessCommand, agentboxRoot, collectAllAgentReservedPorts, getAgentLayout, harnessCapabilities, resolveHarnessCommand } from './agents/index.js';
 export { A as AsyncCommandHandle, C as CommandEvent, a as CommandOptions, b as CommandResult, D as DaytonaProviderOptions, c as DaytonaSandboxOptions, E as E2bProviderOptions, d as E2bSandboxOptions, G as GitCloneOptions, L as LocalDockerProviderOptions, e as LocalDockerSandboxOptions, M as ModalProviderOptions, f as ModalSandboxOptions, S as Sandbox, g as SandboxDescriptor, h as SandboxListOptions, i as SandboxOptions, j as SandboxOptionsBase, k as SandboxOptionsMap, l as SandboxProviderName, m as SandboxRaw, n as SandboxRawMap, o as SandboxResourceSpec, T as TarballEntry, V as VercelGitSource, p as VercelProviderOptions, q as VercelSandboxOptions } from './Sandbox-DcKAU-E3.js';
 export { SandboxAdapter, buildGitCloneCommand } from './sandboxes/index.js';

package/dist/index.js

@@ -15,7 +15,7 @@ import {
   parseHarnessCommandInvocation,
   resolveHarnessCommand,
   skillDirective
-} from "./chunk-REVRD4IY.js";
+} from "./chunk-7PZKMK4U.js";
 import {
   ProviderLogAssembler,
   createNormalizedEvent,
@@ -27,11 +27,11 @@ import {
   Sandbox,
   SandboxAdapter,
   buildGitCloneCommand
-} from "./chunk-76AFK7XN.js";
+} from "./chunk-ZQUE5KT6.js";
 import {
   AGENT_RESERVED_PORTS,
   collectAllAgentReservedPorts
-} from "./chunk-MTQ2S46C.js";
+} from "./chunk-D4LSWCFA.js";
 import "./chunk-NSJM57Z4.js";
 import {
   AgentProvider,

package/dist/sandboxes/index.js

@@ -2,8 +2,8 @@ import {
   Sandbox,
   SandboxAdapter,
   buildGitCloneCommand
-} from "../chunk-76AFK7XN.js";
-import "../chunk-MTQ2S46C.js";
+} from "../chunk-ZQUE5KT6.js";
+import "../chunk-D4LSWCFA.js";
 import "../chunk-NSJM57Z4.js";
 import {
   SandboxProvider

package/dist/{types-DjHP7dFq.d.ts → types-Bhntrw9V.d.ts}

@@ -625,6 +625,20 @@ interface AgentOptionsBase {
      * settles immediately.
      */
     backgroundTaskTimeoutMs?: number;
+    /**
+     * Keep every normalized and raw event in memory for the life of the run so
+     * {@link AgentResult.events} / {@link AgentResult.rawEvents} can replay it
+     * (default `true`).
+     *
+     * Set `false` when the caller consumes the live stream (`for await (const
+     * event of run)` / `run.rawEvents()`) and never reads the arrays off the
+     * result. A long run's transcript — every tool result, every file the agent
+     * read — is otherwise retained twice over and copied once more when the run
+     * settles, which dominates memory for a host running many concurrent runs
+     * in one process. With retention off both arrays settle as `[]`; `text`,
+     * `costData` and every streamed event are unaffected.
+     */
+    retainEvents?: boolean;
     approvalMode?: AgentApprovalMode;
     /** Route questions and plan approvals to the caller even with automatic tool approval. */
     interactiveQuestions?: boolean;
@@ -817,7 +831,9 @@ interface AgentResult {
     text: string;
     isCancelled: boolean;
     error?: string;
+    /** Empty when {@link AgentOptionsBase.retainEvents} is `false`. */
     rawEvents: RawAgentEvent[];
+    /** Empty when {@link AgentOptionsBase.retainEvents} is `false`. */
     events: NormalizedAgentEvent[];
     costData?: AgentCostData | null;
     /**
@@ -970,7 +986,16 @@ interface AgentProviderAdapter<P extends AgentProviderName = AgentProviderName>
      * app-server per run).
      */
     killServer(request: AgentSetupRequest<P>): Promise<void>;
-    execute(request: AgentExecutionRequest<P>, sink: AgentRunSink): Promise<() => Promise<void> | void>;
+    /**
+     * Drive one run to completion. Resolves only once the run has settled
+     * (`sink.complete` / `cancel` / `fail`), so it must not be awaited before
+     * the caller starts reading the sink's streams.
+     *
+     * Cancellation is registered through `sink.setAbort` the moment a
+     * transport exists — not returned from here, which would only ever arrive
+     * after the run was already over.
+     */
+    execute(request: AgentExecutionRequest<P>, sink: AgentRunSink): Promise<void>;
     /**
      * Stateless abort. Dial the in-sandbox provider server, issue the
      * provider's "interrupt the in-flight turn" primitive, close.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "agentbox-sdk",
-  "version": "0.1.519",
+  "version": "0.1.520",
   "description": "Swappable coding agents and sandbox providers for Bun and TypeScript.",
   "license": "MIT",
   "repository": {