@aiden-ade/sandbox-agent 0.1.41 → 0.1.42

package/dist/index.cjs

@@ -13514,6 +13514,16 @@ function resolveBackendRuntimeCommand(backendKind, env = getDaemonCliEnvironment
   cacheResolvedCli(command, resolved);
   return resolved ?? void 0;
 }
+function revalidateBackendRuntimeCommand(backendKind, env = getDaemonCliEnvironment()) {
+  if (!backendKind) return void 0;
+  const command = BACKEND_CLI_COMMANDS[backendKind];
+  if (!command) return void 0;
+  resolvedCliCache.delete(command);
+  const spec = Object.values(PROVIDER_CLI_COMMANDS).find((entry) => entry.command === command);
+  const envOverride = spec ? spec.envVars.map((name) => env[name]?.trim()).find(Boolean) : void 0;
+  if (envOverride) resolvedCliCache.delete(envOverride);
+  return resolveBackendRuntimeCommand(backendKind, env);
+}
 
 // src/computer-readiness.ts
 var import_node_fs5 = require("fs");
@@ -15282,11 +15292,12 @@ var import_os3 = require("os");
 var import_path3 = require("path");
 var import_readline = require("readline");
 var import_child_process2 = require("child_process");
+var import_child_process3 = require("child_process");
 var import_crypto2 = require("crypto");
 var import_fs4 = require("fs");
 var import_os4 = require("os");
 var import_path4 = require("path");
-var import_child_process3 = require("child_process");
+var import_child_process4 = require("child_process");
 var import_util10 = require("util");
 var import_fs5 = require("fs");
 var import_os5 = require("os");
@@ -15876,6 +15887,7 @@ function specForErrorKind(errorKind) {
   const spec = ERROR_SPECS[errorKind];
   return { message: spec.message, errorKind, recoveryClass: spec.recoveryClass };
 }
+var RESUME_SESSION_UNAVAILABLE_MESSAGE = ERROR_SPECS.resume_failed.message;
 function isLikelyProviderAuthError(stderr) {
   const lower = stderr.toLowerCase();
   return lower.includes("api key") || lower.includes("api_key") || lower.includes("invalid api key") || lower.includes("api key invalid") || lower.includes("unauthorized") || lower.includes("authentication failed") || lower.includes("authentication error") || lower.includes("provider settings") || lower.includes("invalid token") || lower.includes("token expired");
@@ -16031,6 +16043,54 @@ function spawnCli(command, args, context) {
     detached: process.platform !== "win32"
   });
 }
+function isAlreadyDead(error) {
+  return error?.code === "ESRCH";
+}
+function killProcessTree(pid, options = {}) {
+  const { signal = "SIGTERM", child, onError } = options;
+  if (typeof pid !== "number" || !Number.isInteger(pid) || pid <= 0) {
+    if (child) {
+      try {
+        child.kill(signal);
+      } catch (err) {
+        if (!isAlreadyDead(err)) onError?.("fallback", err);
+      }
+    }
+    return;
+  }
+  if (process.platform === "win32") {
+    try {
+      const killer = (0, import_child_process3.spawn)("taskkill", ["/pid", String(pid), "/t", "/f"], {
+        stdio: "ignore",
+        windowsHide: true
+      });
+      killer.once("error", (err) => {
+        if (child) {
+          try {
+            child.kill();
+          } catch {
+          }
+        }
+        onError?.("taskkill", err);
+      });
+    } catch (err) {
+      onError?.("taskkill", err);
+    }
+    return;
+  }
+  try {
+    process.kill(-pid, signal);
+  } catch (groupErr) {
+    if (!isAlreadyDead(groupErr)) onError?.("group", groupErr);
+    if (child) {
+      try {
+        child.kill(signal);
+      } catch (fallbackErr) {
+        if (!isAlreadyDead(fallbackErr)) onError?.("fallback", fallbackErr);
+      }
+    }
+  }
+}
 function emitSessionNotice(presenter, kind, message) {
   if (presenter.onNotice) {
     void presenter.onNotice(kind, message);
@@ -16095,6 +16155,12 @@ function formatCliSpawnError(command, error) {
   }
   return error;
 }
+function waitForSpawnOutcome(child) {
+  return new Promise((resolve22) => {
+    child.once("spawn", () => resolve22(null));
+    child.once("error", (error) => resolve22(error));
+  });
+}
 function createGenericCliBackend(options) {
   return {
     kind: options.kind,
@@ -16108,7 +16174,23 @@ function createGenericCliBackend(options) {
       };
       const prompt = options.augmentPrompt?.(context) ?? (context.config.mode === "plan" ? buildPlanModePrefix(buildPromptWithSystem(context.config, context.promptText)) : buildPromptWithSystem(context.config, context.promptText));
       const args = options.buildArgs?.(context, prompt) ?? options.args.map((arg) => arg === "{{prompt}}" ? prompt : arg);
-      const child = spawnCli(options.command, args, context);
+      let command = options.command;
+      let child = spawnCli(command, args, context);
+      if (context.revalidateCommand) {
+        const spawnError = await waitForSpawnOutcome(child);
+        if (spawnError) {
+          const fresh = spawnError.code === "ENOENT" ? context.revalidateCommand()?.trim() || null : null;
+          if (fresh && fresh !== command) {
+            console.warn(
+              `[${options.kind}] Spawn failed ENOENT for "${command}" \u2014 re-resolved to "${fresh}", retrying once`
+            );
+            command = fresh;
+            child = spawnCli(command, args, context);
+          } else {
+            throw formatCliSpawnError(command, spawnError);
+          }
+        }
+      }
       state.process = child;
       state.lastRawOutputAtMs = Date.now();
       context.onProcessSpawned?.(child);
@@ -16119,8 +16201,8 @@ function createGenericCliBackend(options) {
       const safeArgs = args.map(
         (a, i) => i > 0 && args[i - 1] === "--system-prompt" ? `"<system-prompt ${a.length} chars>"` : a
       );
-      console.info(`[${options.kind}] Spawning: ${options.command} ${safeArgs.join(" ")}`);
-      presenter.recordRawTranscript?.("system", `${options.command} ${args.join(" ")}`, {
+      console.info(`[${options.kind}] Spawning: ${command} ${safeArgs.join(" ")}`);
+      presenter.recordRawTranscript?.("system", `${command} ${args.join(" ")}`, {
         backendKind: options.kind
       });
       child.stdin.on("error", (err) => {
@@ -16152,14 +16234,7 @@ function createGenericCliBackend(options) {
           console.warn(
             `[${options.kind}] No stdout within ${options.startupTimeoutMs}ms of spawn \u2014 killing hung process`
           );
-          try {
-            if (process.platform === "win32") {
-              child.kill("SIGKILL");
-            } else if (child.pid) {
-              process.kill(-child.pid, "SIGKILL");
-            }
-          } catch {
-          }
+          killProcessTree(child.pid, { signal: "SIGKILL", child });
         }, options.startupTimeoutMs);
       }
       stdoutRl.on("line", (line) => {
@@ -16202,16 +16277,7 @@ function createGenericCliBackend(options) {
           });
           child.on("exit", (code) => resolve22(code ?? 0));
           const onAbort = () => {
-            const pid = child.pid;
-            if (!pid) return;
-            try {
-              process.kill(-pid, "SIGTERM");
-            } catch {
-              try {
-                child.kill("SIGTERM");
-              } catch {
-              }
-            }
+            killProcessTree(child.pid, { signal: "SIGTERM", child });
           };
           context.abortController.signal.addEventListener("abort", onAbort, { once: true });
         });
@@ -16247,14 +16313,7 @@ function createGenericCliBackend(options) {
             console.warn(
               `[${options.kind}] Idle timeout (${IDLE_MS}ms) waiting for ${idleTimeoutReason === "background_task" ? "a background task terminal event" : "a pending user answer"} \u2014 force-killing orphaned process`
             );
-            try {
-              if (process.platform === "win32") {
-                child.kill("SIGKILL");
-              } else if (child.pid) {
-                process.kill(-child.pid, "SIGKILL");
-              }
-            } catch {
-            }
+            killProcessTree(child.pid, { signal: "SIGKILL", child });
             exitCode = await Promise.race([
               exitPromise,
               new Promise((resolve22) => setTimeout(() => resolve22(1), 5e3))
@@ -16270,14 +16329,7 @@ function createGenericCliBackend(options) {
               console.warn(
                 `[${options.kind}] Process still alive ${GRACE_MS}ms after result event, killing`
               );
-              try {
-                if (process.platform === "win32") {
-                  child.kill("SIGKILL");
-                } else if (child.pid) {
-                  process.kill(-child.pid, "SIGKILL");
-                }
-              } catch {
-              }
+              killProcessTree(child.pid, { signal: "SIGKILL", child });
             }
           } else {
             exitCode = raceResult;
@@ -16440,6 +16492,9 @@ function createGenericCliBackend(options) {
           classifiedErrorKind = detailed.errorKind;
           classifiedRecoveryClass = detailed.recoveryClass;
         }
+        if (classifiedErrorKind === "resume_failed" && !requestedResumeId) {
+          classifiedErrorKind = void 0;
+        }
       }
       return {
         success: !failed,
@@ -16448,7 +16503,11 @@ function createGenericCliBackend(options) {
         planFilesCreated: [],
         iterations: Math.max(state.iterations, 1),
         error: classifiedError,
-        ...classifiedErrorKind ? { errorKind: classifiedErrorKind } : {},
+        // Surface every recognized, machine-actionable kind (model_mismatch,
+        // resume_failed, auth_*, …). The catch-all unknown_cli_error carries no routing
+        // signal — nothing downstream keys on it — so leave errorKind absent for it,
+        // keeping an unrelated failure untagged even when a resume id was set.
+        ...classifiedErrorKind && classifiedErrorKind !== "unknown_cli_error" ? { errorKind: classifiedErrorKind } : {},
         ...classifiedRecoveryClass ? { recoveryClass: classifiedRecoveryClass } : {},
         providerSessionId: state.runtimeSessionId,
         runtimeSessionId: state.runtimeSessionId,
@@ -16906,7 +16965,7 @@ function claudeSessionLogExists(input) {
   if (!(0, import_fs4.existsSync)(projectDir)) return false;
   return (0, import_fs4.existsSync)((0, import_path4.join)(projectDir, `${input.sessionId}.jsonl`));
 }
-var execFileAsync = (0, import_util10.promisify)(import_child_process3.execFile);
+var execFileAsync = (0, import_util10.promisify)(import_child_process4.execFile);
 var flagSupportCache = /* @__PURE__ */ new Map();
 function helpAdvertisesFlag(helpText, flag) {
   return helpText.includes(flag);
@@ -17641,18 +17700,7 @@ function armCodexExitGraceKill(state) {
     console.warn(
       "[codex_app_server] Process did not exit after turn completion; killing process group"
     );
-    const pid = child.pid;
-    if (pid) {
-      try {
-        process.kill(-pid, "SIGTERM");
-        return;
-      } catch {
-      }
-    }
-    try {
-      child.kill("SIGTERM");
-    } catch {
-    }
+    killProcessTree(child.pid, { signal: "SIGTERM", child });
   }, CODEX_EXIT_GRACE_MS);
   timer.unref?.();
   state.exitGraceKillTimer = timer;
@@ -19231,17 +19279,7 @@ function handleOpencodeStructuredEvent(parsed, context, state) {
         void presenter.onError(message);
         state.error = message;
       }
-      const pid = state.process?.pid;
-      if (pid) {
-        try {
-          process.kill(-pid, "SIGTERM");
-        } catch {
-          try {
-            state.process?.kill("SIGTERM");
-          } catch {
-          }
-        }
-      }
+      killProcessTree(state.process?.pid, { signal: "SIGTERM", child: state.process });
       return true;
     }
     default:
@@ -19791,6 +19829,8 @@ Prefer relative paths and keep your work scoped to this project.`
       runtimeCommand || "generic-cli",
       runtimeArgs
     );
+    const revalidateRuntimeCommand = this.revalidateRuntimeCommand;
+    const revalidateCommand = revalidateRuntimeCommand ? () => revalidateRuntimeCommand.call(this) : void 0;
     let lastResult = null;
     for (let attempt = 0; attempt <= _BaseMachineAgent.MAX_RETRIES; attempt++) {
       if (this.abortController?.signal.aborted) break;
@@ -19801,7 +19841,8 @@ Prefer relative paths and keep your work scoped to this project.`
         cwd,
         env,
         promptText,
-        onProcessSpawned: this.getOnProcessSpawned()
+        onProcessSpawned: this.getOnProcessSpawned(),
+        revalidateCommand
       });
       lastResult = {
         ...result,
@@ -19851,7 +19892,8 @@ ${runtimeConfig.task}` } : {}
           cwd,
           env,
           promptText: fallbackContext ? this.buildPromptText(runtimeConfig) : promptText,
-          onProcessSpawned: this.getOnProcessSpawned()
+          onProcessSpawned: this.getOnProcessSpawned(),
+          revalidateCommand
         });
         lastResult = {
           ...retryResult,
@@ -20271,38 +20313,32 @@ var CoreAgent = class _CoreAgent extends BaseMachineAgent {
     const pid = child.pid;
     if (!pid) return false;
     console.info("[CoreAgent] Killing agent process", { pid });
-    try {
-      if (process.platform === "win32") {
-        child.kill("SIGTERM");
-      } else {
-        process.kill(-pid, "SIGTERM");
-      }
-    } catch (err) {
-      if (err.code !== "ESRCH") {
-        console.warn("[CoreAgent] SIGTERM failed", { pid, err });
-      }
-      try {
-        child.kill("SIGTERM");
-      } catch {
-      }
-    }
+    killProcessTree(pid, {
+      signal: "SIGTERM",
+      child,
+      onError: (stage, err) => console.warn(`[CoreAgent] SIGTERM failed (${stage})`, { pid, err })
+    });
     setTimeout(() => {
       if (child.killed || child.exitCode !== null) return;
       console.warn("[CoreAgent] Escalating to SIGKILL", { pid });
-      try {
-        if (process.platform === "win32") {
-          child.kill("SIGKILL");
-        } else {
-          process.kill(-pid, "SIGKILL");
-        }
-      } catch (err) {
-        if (err.code !== "ESRCH") {
-          console.warn("[CoreAgent] SIGKILL failed", { pid, err });
-        }
-      }
+      killProcessTree(pid, {
+        signal: "SIGKILL",
+        child,
+        onError: (stage, err) => console.warn(`[CoreAgent] SIGKILL failed (${stage})`, { pid, err })
+      });
     }, _CoreAgent.SIGKILL_DELAY_MS);
     return true;
   }
+  /**
+   * Re-resolve this agent's CLI executable from scratch when a spawn fails
+   * ENOENT. The daemon caches resolved CLI paths, so a provider binary that was
+   * moved or reinstalled mid-session would otherwise fail every turn with
+   * "CLI command not found" until the daemon restarts. Invalidating the cache
+   * and re-resolving once lets the current run recover against the new location.
+   */
+  revalidateRuntimeCommand() {
+    return revalidateBackendRuntimeCommand(this.runtime.backendKind) ?? null;
+  }
   // ── Interactive tool response (WS relay → stdin) ───────────────────────
   /**
    * Send a tool_result to the CLI's stdin (stream-json format).
@@ -20655,6 +20691,8 @@ var EncryptedEventOutbox = class {
   inFlightEventId = null;
   ackTimer = null;
   retryTimer = null;
+  /** Latch so a persistent disk failure logs once per streak, not per event. */
+  persistFailureLogged = false;
   key;
   maxEncryptedBytes;
   maxIntermediateEventsPerRun;
@@ -20858,30 +20896,42 @@ var EncryptedEventOutbox = class {
     }
   }
   persist() {
-    if (this.entries.length === 0) {
-      (0, import_node_fs6.rmSync)(this.path, { force: true });
-      return;
-    }
-    (0, import_node_fs6.mkdirSync)((0, import_node_path5.dirname)(this.path), { recursive: true, mode: 448 });
-    const iv = (0, import_node_crypto.randomBytes)(12);
-    const cipher = (0, import_node_crypto.createCipheriv)("aes-256-gcm", this.key, iv);
-    const ciphertext = Buffer.concat([
-      cipher.update(JSON.stringify({ entries: this.entries }), "utf8"),
-      cipher.final()
-    ]);
-    const envelope = {
-      version: OUTBOX_VERSION,
-      iv: iv.toString("base64url"),
-      authTag: cipher.getAuthTag().toString("base64url"),
-      ciphertext: ciphertext.toString("base64url")
-    };
-    const pendingPath = `${this.path}.pending-${process.pid}-${(0, import_node_crypto.randomBytes)(6).toString("hex")}`;
     try {
-      (0, import_node_fs6.writeFileSync)(pendingPath, JSON.stringify(envelope), { mode: 384 });
-      (0, import_node_fs6.chmodSync)(pendingPath, 384);
-      (0, import_node_fs6.renameSync)(pendingPath, this.path);
-    } finally {
-      (0, import_node_fs6.rmSync)(pendingPath, { force: true });
+      if (this.entries.length === 0) {
+        (0, import_node_fs6.rmSync)(this.path, { force: true });
+        this.persistFailureLogged = false;
+        return;
+      }
+      (0, import_node_fs6.mkdirSync)((0, import_node_path5.dirname)(this.path), { recursive: true, mode: 448 });
+      const iv = (0, import_node_crypto.randomBytes)(12);
+      const cipher = (0, import_node_crypto.createCipheriv)("aes-256-gcm", this.key, iv);
+      const ciphertext = Buffer.concat([
+        cipher.update(JSON.stringify({ entries: this.entries }), "utf8"),
+        cipher.final()
+      ]);
+      const envelope = {
+        version: OUTBOX_VERSION,
+        iv: iv.toString("base64url"),
+        authTag: cipher.getAuthTag().toString("base64url"),
+        ciphertext: ciphertext.toString("base64url")
+      };
+      const pendingPath = `${this.path}.pending-${process.pid}-${(0, import_node_crypto.randomBytes)(6).toString("hex")}`;
+      try {
+        (0, import_node_fs6.writeFileSync)(pendingPath, JSON.stringify(envelope), { mode: 384 });
+        (0, import_node_fs6.chmodSync)(pendingPath, 384);
+        (0, import_node_fs6.renameSync)(pendingPath, this.path);
+      } finally {
+        (0, import_node_fs6.rmSync)(pendingPath, { force: true });
+      }
+      this.persistFailureLogged = false;
+    } catch (error) {
+      if (!this.persistFailureLogged) {
+        this.persistFailureLogged = true;
+        const detail = error instanceof Error ? error.message : String(error);
+        this.pushLog(
+          `event_outbox_persist_failed error=${detail} pending_entries=${this.entries.length}`
+        );
+      }
     }
   }
 };
@@ -22527,7 +22577,7 @@ var RunStartGate = class {
 };
 
 // src/version.ts
-var AGENT_VERSION = "0.1.41";
+var AGENT_VERSION = "0.1.42";
 
 // src/workspace-relocation.ts
 var import_node_child_process3 = require("child_process");

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@aiden-ade/sandbox-agent",
-  "version": "0.1.41",
+  "version": "0.1.42",
   "type": "module",
   "description": "Alan agent runtime — cloud sandbox and local daemon (alan-agent CLI)",
   "bin": {