@evident-ai/cli 3.5.2-dev.ffa823b → 3.6.1-dev.8d3e144

package/dist/index.js

@@ -1293,7 +1293,11 @@ var FORWARD_FAILURE_REASON_HEADER = "X-Evident-Failure-Reason";
 function log(level, event, fields) {
   const method = level === "debug" ? "log" : level;
   try {
-    console[method]("[evident]", JSON.stringify({ level, event, ...fields }));
+    const payload = Object.fromEntries(
+      Object.entries({ level, event, ...fields }).filter(([, value]) => value !== void 0)
+    );
+    const secondArgument = globalThis.navigator?.userAgent === "Cloudflare-Workers" ? payload : JSON.stringify(payload);
+    console[method]("[evident]", secondArgument);
   } catch (err) {
     console.error(
       "[evident] log_serialize_failed",
@@ -1322,7 +1326,7 @@ import ora4 from "ora";
 import { select as select4 } from "@inquirer/prompts";
 
 // src/lib/telemetry.ts
-var CLI_VERSION = (true ? "3.5.2-dev.ffa823b" : void 0) ?? process.env.npm_package_version ?? "unknown";
+var CLI_VERSION = (true ? "3.6.1-dev.8d3e144" : void 0) ?? process.env.npm_package_version ?? "unknown";
 function getCliVersion() {
   return CLI_VERSION;
 }
@@ -3518,19 +3522,32 @@ function isAbortedTerminalReply(messages, userMessageId) {
   }
   return false;
 }
+var CLAUDE_AUTH_EXPIRY_MESSAGES = Object.freeze([
+  "Claude Code credentials are unavailable or expired. Run `claude` to refresh them."
+]);
+function extractReplyErrorShape(reply) {
+  const error2 = errorOf(reply);
+  if (error2 == null || typeof error2 !== "object") {
+    return { name: null, statusCode: null };
+  }
+  const e = error2;
+  const name = typeof e.name === "string" ? e.name.slice(0, 100) : null;
+  const data = e.data;
+  const statusCode = typeof data?.statusCode === "number" && Number.isFinite(data.statusCode) ? data.statusCode : null;
+  return { name, statusCode };
+}
 function classifyReplyAuthError(reply) {
   const error2 = errorOf(reply);
   if (error2 == null || typeof error2 !== "object") return null;
   const e = error2;
   const replyProviderId = reply?.info?.providerID ?? null;
   const replyModelId = reply?.info?.modelID ?? null;
+  const data = e.data;
   if (e.name === "ProviderAuthError") {
-    const data = e.data;
     const providerId = typeof data?.providerID === "string" && data.providerID || replyProviderId;
     return { kind: "model_auth", providerId, modelId: replyModelId, reason: "missing" };
   }
   if (e.name === "APIError") {
-    const data = e.data;
     const statusCode = data?.statusCode;
     if (statusCode === 401 || statusCode === 403) {
       return {
@@ -3541,11 +3558,23 @@ function classifyReplyAuthError(reply) {
       };
     }
   }
+  const errorMessage3 = typeof data?.message === "string" ? data.message : typeof e.message === "string" ? e.message : null;
+  if (errorMessage3 !== null && CLAUDE_AUTH_EXPIRY_MESSAGES.some((knownMessage) => errorMessage3.includes(knownMessage))) {
+    return {
+      kind: "model_auth",
+      providerId: typeof data?.providerID === "string" && data.providerID || replyProviderId,
+      modelId: replyModelId,
+      reason: "rejected"
+    };
+  }
   return null;
 }
 function messageFailure(messages, userMessageId) {
   return classifyReplyAuthError(findLastAssistantReplyFor(messages, userMessageId));
 }
+function isProviderModelNotFoundError(reason) {
+  return reason.includes("Model not found:");
+}
 function findLatestSubagentAuthOutcome(messages, sinceMs) {
   if (!messages || messages.length === 0) return null;
   for (let i = messages.length - 1; i >= 0; i--) {
@@ -3754,23 +3783,49 @@ async function readSessionErrorStream(port, options, client) {
     if (reader) void reader.cancel().catch(() => void 0);
   }
 }
-async function reloadProviderCache(port, client) {
-  if (client?.version === "v2") return;
+async function disposeAndReloadProviderConfig(port, client) {
+  if (client?.version === "v2") return { outcome: "skipped_v2" };
   try {
-    const res = await requestWithClient(port, client, "/config", {
-      method: "PATCH",
-      headers: { "Content-Type": "application/json" },
-      body: JSON.stringify({})
-    });
+    const res = await requestWithClient(
+      port,
+      client,
+      "/config",
+      {
+        method: "PATCH",
+        headers: { "Content-Type": "application/json" },
+        body: JSON.stringify({})
+      },
+      { allowStatuses: ALL_HTTP_STATUSES }
+    );
     if (!res.ok) {
       console.error(
-        `[reloadProviderCache] PATCH /config returned HTTP ${res.status} (port ${port})`
+        `[disposeAndReloadProviderConfig] PATCH /config returned HTTP ${res.status} (port ${port})`
       );
+      return { outcome: "failed", status: res.status, detail: null };
     }
+    return { outcome: "reloaded" };
   } catch (err) {
     console.error(
-      `[reloadProviderCache] PATCH /config failed (port ${port}): ${err instanceof Error ? err.message : String(err)}`
+      `[disposeAndReloadProviderConfig] PATCH /config failed (port ${port}): ${err instanceof Error ? err.message : String(err)}`
     );
+    return {
+      outcome: "failed",
+      status: null,
+      detail: err instanceof Error ? err.message : String(err)
+    };
+  }
+}
+async function probeOpenCodeRequestReadiness(port, client) {
+  try {
+    const res = await requestWithClient(port, client, "/config", void 0, {
+      allowStatuses: ALL_HTTP_STATUSES
+    });
+    const body = await res.text();
+    return { ok: res.ok, status: res.status, bodyEmpty: body.trim().length === 0, detail: null };
+  } catch (err) {
+    const detail = err instanceof Error ? err.message : String(err);
+    console.error(`[probeOpenCodeRequestReadiness] GET /config failed (port ${port}): ${detail}`);
+    return { ok: false, status: null, bodyEmpty: null, detail };
   }
 }
 
@@ -7537,7 +7592,7 @@ var SESSION_ERROR_BUFFER_TTL_MS = SESSION_ERROR_STREAM_HEALTHY_MS;
 var MAX_BUFFERED_SESSION_ERRORS = 256;
 var SESSION_ERROR_RETENTION_TTL_MS = 6e4;
 var SESSION_STORE_UNWRITABLE_ERROR = "The runner session database could not be written, so this turn did not finish. Nothing was lost from the queue. Clear disk space and retry.";
-var SESSION_ERROR_RETAINED_DETAIL = "The runner's session database could not be written, so this turn stopped part-way. Runner-observed session error: ";
+var SESSION_ERROR_RETAINED_DETAIL = "Runner-observed OpenCode session error: ";
 var SESSION_ERROR_OBSERVED_CAUSE = "Runner-observed OpenCode session-store cause: ";
 var DEFAULT_PAUSED_POLL_INTERVAL_MS = 2e3;
 var DEFAULT_PAUSED_MAX_WAIT_MS = 10 * 60 * 1e3;
@@ -7661,12 +7716,15 @@ var ChannelDriver = class _ChannelDriver {
   /**
    * Per-conversation throttle state for the #183 recurrence warning (#1618
    * WI-4) — see `reportWedgedConversation`'s doc comment for why this exists.
-   * `firstWedgedAt` anchors `stuck_for_ms`; `lastWarnedAt` throttles both the
-   * log line and the `dispatch_wedged` signal to at most once per
-   * `wedgeWarningIntervalMs`; `consecutiveTicks` is reported in the log text
-   * so the operator sees magnitude, not repetition. Cleared the moment the
-   * conversation dispatches anything (a fresh wedge, if it recurs, is a new
-   * incident). Bounded FIFO, mirroring `supersededSessions`
+   * `firstWedgedAt` anchors `stuck_for_ms`; `warnAnchorAt` is the throttle
+   * anchor from first detection through the first warning, then from each
+   * warning onward; `consecutiveTicks` is reported in the log text so the
+   * operator sees magnitude, not repetition. Cleared the moment the
+   * conversation dispatches anything, or when it drops off the pending list
+   * (`runDrain`'s prune, the dominant real-world path — a wedge normally ends
+   * with its message completing, not a fresh dispatch on the same
+   * conversation) — either way a fresh wedge, if it recurs, is a new
+   * incident. Bounded FIFO, mirroring `supersededSessions`
    * (`MAX_WEDGED_CONVERSATIONS`).
    */
   wedgeWarnings = /* @__PURE__ */ new Map();
@@ -7896,6 +7954,7 @@ var ChannelDriver = class _ChannelDriver {
   subagentInvocationPrefetches = /* @__PURE__ */ new Map();
   /** Serialises drains so a reconnect during a drain doesn't double-process. */
   draining = false;
+  pendingProviderRegistryRepair = null;
   /**
    * Serialises runner-file syncs (#559) so the ~2s poll tick and a concurrent
    * drain ping don't download, write and ack the same file twice.
@@ -8081,8 +8140,33 @@ var ChannelDriver = class _ChannelDriver {
   async runDrain() {
     let dispatched = 0;
     try {
+      const pendingRepair = this.pendingProviderRegistryRepair;
+      if (pendingRepair && this.protectedSessionIds().size === 0) {
+        try {
+          await this.issueProviderRegistryReload(
+            pendingRepair.conversationId,
+            pendingRepair.messageId
+          );
+        } catch (err) {
+          this.log({
+            level: "error",
+            message: `Provider registry repair adoption failed for message ${pendingRepair.messageId.slice(0, 8)}: ${err instanceof Error ? err.message : String(err)}`,
+            conversation_id: pendingRepair.conversationId,
+            message_id: pendingRepair.messageId
+          });
+        }
+        return 0;
+      }
       this.pendingMessageIds.clear();
       const conversations = await this.getPendingConversations();
+      if (this.wedgeWarnings.size > 0) {
+        const pendingConversationIds = new Set(conversations.map(({ id }) => id));
+        for (const conversationId of this.wedgeWarnings.keys()) {
+          if (!pendingConversationIds.has(conversationId)) {
+            this.wedgeWarnings.delete(conversationId);
+          }
+        }
+      }
       if (this.recycleRequestedFlag) {
         this.stop();
       }
@@ -8619,14 +8703,13 @@ var ChannelDriver = class _ChannelDriver {
    * skip-because-already-`dispatched`, the exact signature of a message stuck
    * acknowledged-but-never-worked.
    *
-   * Unthrottled, this fired every ~6s drain tick for as long as a wedge lasted
-   * (52,843 occurrences observed in one incident) — burning the GLOBAL
-   * 30-events/60s `runner-activity-telemetry.ts` budget that was itself
-   * suppressing the diagnostics needed to debug the wedge. The `warn` log (and
-   * the `dispatch_wedged` signal once the wedge has persisted past the same
-   * interval) fire at most once per `wedgeWarningIntervalMs` per conversation,
-   * naming the consecutive-tick count so the operator sees magnitude rather
-   * than repetition.
+   * The first detection starts the clock and emits a local `debug` entry. The
+   * `warn` log and paired `dispatch_wedged` signal fire once the wedge has
+   * lasted `wedgeWarningIntervalMs`, then at most once per interval, naming the
+   * consecutive-tick count so the operator sees magnitude rather than repetition.
+   * A watcher stall that the loop-liveness watchdog recovers is expected and is
+   * already reported by its `watcher_recovered` signal, so it does not reach an
+   * operator through this warning.
    *
    * Deliberately does NOT trigger a release: WI-1's `reconcileWatchers` runs
    * unconditionally on this same tick and is already recovering anything it
@@ -8638,36 +8721,41 @@ var ChannelDriver = class _ChannelDriver {
     const existing = this.wedgeWarnings.get(conv.id);
     const firstWedgedAt = existing?.firstWedgedAt ?? now;
     const consecutiveTicks = (existing?.consecutiveTicks ?? 0) + 1;
-    const dueForWarn = !existing || now - existing.lastWarnedAt >= this.wedgeWarningIntervalMs;
-    if (!dueForWarn) {
-      this.wedgeWarnings.delete(conv.id);
-      this.wedgeWarnings.set(conv.id, {
-        firstWedgedAt,
-        lastWarnedAt: existing.lastWarnedAt,
-        consecutiveTicks
+    const warnAnchorAt = existing?.warnAnchorAt ?? now;
+    const dueForWarn = now - warnAnchorAt >= this.wedgeWarningIntervalMs;
+    if (!existing) {
+      this.log({
+        level: "debug",
+        message: `Detected a possible wedge for conversation ${conv.id.slice(0, 8)} with ${messages.length} pending message(s); it will be reported only if it outlasts the watchdog's recovery.`,
+        conversation_id: conv.id
+      });
+    }
+    let untracked = 0;
+    if (dueForWarn) {
+      const stuckForMs = now - firstWedgedAt;
+      untracked = this.countUntrackedIds(messages);
+      this.log({
+        level: "warn",
+        message: `Conversation ${conv.id.slice(0, 8)} has ${messages.length} pending message(s) but ALL are already marked dispatched locally (in-flight set: ${this.dispatched.size}) \u2014 none sent to OpenCode for ${consecutiveTicks} consecutive tick(s) now (${Math.round(stuckForMs / 1e3)}s stuck). ` + (untracked > 0 ? `${untracked} of these id(s) are tracked by NO watcher \u2014 the dispatched/in-flight pairing invariant is violated for this conversation, which will NOT self-heal and needs a runner restart.` : `A watcher is tracking this work; the loop-liveness watchdog is already recovering it.`),
+        conversation_id: conv.id
       });
-      return;
     }
-    const stuckForMs = now - firstWedgedAt;
-    const untracked = this.countUntrackedIds(messages);
-    this.log({
-      level: "warn",
-      message: `Conversation ${conv.id.slice(0, 8)} has ${messages.length} pending message(s) but ALL are already marked dispatched locally (in-flight set: ${this.dispatched.size}) \u2014 none sent to OpenCode for ${consecutiveTicks} consecutive tick(s) now (${stuckForMs}ms stuck). ` + (untracked > 0 ? `${untracked} of these id(s) are tracked by NO watcher \u2014 the dispatched/in-flight pairing invariant is violated for this conversation, which will NOT self-heal and needs a runner restart.` : `A watcher is tracking this work; the loop-liveness watchdog is already recovering it.`),
-      conversation_id: conv.id
-    });
     this.wedgeWarnings.delete(conv.id);
-    this.wedgeWarnings.set(conv.id, { firstWedgedAt, lastWarnedAt: now, consecutiveTicks });
+    this.wedgeWarnings.set(conv.id, {
+      firstWedgedAt,
+      warnAnchorAt: dueForWarn ? now : warnAnchorAt,
+      consecutiveTicks
+    });
     while (this.wedgeWarnings.size > MAX_WEDGED_CONVERSATIONS) {
       const oldest = this.wedgeWarnings.keys().next().value;
       if (oldest === void 0) break;
       this.wedgeWarnings.delete(oldest);
     }
-    if (stuckForMs >= this.wedgeWarningIntervalMs) {
-      void this.postSignal(conv.id, messages[0].id, "dispatch_wedged", {
-        stuck_for_ms: stuckForMs,
-        untracked
-      });
-    }
+    if (!dueForWarn) return;
+    void this.postSignal(conv.id, messages[0].id, "dispatch_wedged", {
+      stuck_for_ms: now - firstWedgedAt,
+      untracked
+    });
   }
   /**
    * How many of `messages`' ids are tracked by NO watcher's `inFlight` (#1618
@@ -9933,14 +10021,40 @@ var ChannelDriver = class _ChannelDriver {
         conversation_id: watcher.conv.id,
         message_id: inFlight.evidentMessageId
       });
-      await this.markFailed(
-        watcher.conv.id,
-        inFlight.evidentMessageId,
-        event.sessionId,
-        `OpenCode could not run this turn: ${event.reason}`
-      );
+      const isMissingProvider = isProviderModelNotFoundError(event.reason);
+      if (isMissingProvider) {
+        const dispatchedModel = inFlight.message.opencode_model;
+        const separator = dispatchedModel?.indexOf("/") ?? -1;
+        const hasProviderAndModel = dispatchedModel !== null && separator > 0 && separator < dispatchedModel.length - 1;
+        await this.markFailed(
+          watcher.conv.id,
+          inFlight.evidentMessageId,
+          event.sessionId,
+          `OpenCode could not run this turn: ${event.reason}`,
+          void 0,
+          {
+            classification: {
+              kind: "model_auth",
+              providerId: hasProviderAndModel ? dispatchedModel.slice(0, separator) : null,
+              modelId: hasProviderAndModel ? dispatchedModel.slice(separator + 1) : null,
+              reason: "missing"
+            },
+            errorShape: { name: null, statusCode: null }
+          }
+        );
+      } else {
+        await this.markFailed(
+          watcher.conv.id,
+          inFlight.evidentMessageId,
+          event.sessionId,
+          `OpenCode could not run this turn: ${event.reason}`
+        );
+      }
       inFlight.done = true;
       this.removeInFlight(watcher, inFlight.evidentMessageId);
+      if (isMissingProvider) {
+        await this.repairProviderRegistry(watcher.conv.id, inFlight.evidentMessageId);
+      }
     } catch (err) {
       if (err instanceof ChannelAuthError) {
         this.log({
@@ -9961,6 +10075,44 @@ var ChannelDriver = class _ChannelDriver {
       this.sessionErrorHandled.delete(inFlight.evidentMessageId);
     }
   }
+  async repairProviderRegistry(conversationId, messageId) {
+    if (this.isV2) {
+      void this.postSignal(conversationId, messageId, "provider_registry_reload_unsupported");
+      return;
+    }
+    const activeSessions = this.protectedSessionIds().size;
+    if (activeSessions > 0 || this.draining) {
+      this.pendingProviderRegistryRepair = { conversationId, messageId };
+      void this.postSignal(conversationId, messageId, "provider_registry_repair_deferred", {
+        active_sessions: activeSessions
+      });
+      this.log({
+        level: "warn",
+        message: `Provider registry repair deferred until a lifecycle-safe point; ${activeSessions} active session(s) protected`,
+        conversation_id: conversationId,
+        message_id: messageId
+      });
+      return;
+    }
+    await this.issueProviderRegistryReload(conversationId, messageId);
+  }
+  async issueProviderRegistryReload(conversationId, messageId) {
+    this.pendingProviderRegistryRepair = null;
+    const result = await disposeAndReloadProviderConfig(this.port, this.openCodeClient);
+    if (result.outcome === "failed") {
+      const failure = result.status === null ? `transport failure${result.detail ? `: ${result.detail}` : ""}` : `HTTP ${result.status}`;
+      this.log({
+        level: "warn",
+        message: `Provider registry reload failed (${failure})`,
+        conversation_id: conversationId,
+        message_id: messageId
+      });
+    }
+    const repairOutcome = result.outcome === "reloaded" ? "reloaded" : "reload_failed";
+    void this.postSignal(conversationId, messageId, "provider_registry_reload_issued", {
+      repair_outcome: repairOutcome
+    });
+  }
   /**
    * The per-session polling loop (WI-3). Once per tick it:
    *   1. polls `GET /session/:id/message` once and, per in-flight message,
@@ -10152,7 +10304,7 @@ var ChannelDriver = class _ChannelDriver {
       if (!inFlight.done) {
         const messageErrorText = messageError(messages, inFlight.opencodeMessageId);
         const retainedReason = this.readRetainedSessionError(sessionId);
-        const retainedError = retainedReason === null ? null : `${SESSION_ERROR_RETAINED_DETAIL}${retainedReason}`;
+        const retainedError = retainedReason === null || messageErrorText !== null && messageErrorText.includes(retainedReason) ? null : `${SESSION_ERROR_RETAINED_DETAIL}${retainedReason}`;
         const observedCause = this.sessionStoreHealth?.readCause(sessionId);
         const causeError = observedCause === null || observedCause === void 0 ? null : `${SESSION_ERROR_OBSERVED_CAUSE}${observedCause}`;
         const storeError = this.sessionStoreHealth?.isUnwritable() ? SESSION_STORE_UNWRITABLE_ERROR : null;
@@ -10648,7 +10800,8 @@ var ChannelDriver = class _ChannelDriver {
         usage,
         void 0,
         usageAgentName,
-        subagentInvocations
+        subagentInvocations,
+        true
       );
     } catch (err) {
       if (err instanceof ChannelAuthError) throw err;
@@ -12494,7 +12647,7 @@ var ChannelDriver = class _ChannelDriver {
    *     exists but is wedged, so the next attempt must get a fresh one
    *     instead of reusing it — see WI-1's server-side null-clearing PATCH).
    */
-  async markFailed(conversationId, messageId, sessionId, error2, usage, failure, usageAgentName, subagentInvocations) {
+  async markFailed(conversationId, messageId, sessionId, error2, usage, failure, usageAgentName, subagentInvocations, abandonedWithoutAnswer) {
     const body = { status: "failed" };
     if (sessionId === null) {
       body.opencode_session_id = null;
@@ -12507,12 +12660,19 @@ var ChannelDriver = class _ChannelDriver {
     if (subagentInvocations && subagentInvocations.length > 0) {
       body.subagent_invocations = subagentInvocations;
     }
-    if (failure) {
-      body.failure_kind = failure.kind;
-      body.failure_provider_id = failure.providerId;
-      body.failure_model_id = failure.modelId;
-      body.failure_reason = failure.reason;
+    if (failure && failure.errorShape.name !== null) {
+      body.failure_error_name = failure.errorShape.name;
+    }
+    if (failure && failure.errorShape.statusCode !== null) {
+      body.failure_status_code = failure.errorShape.statusCode;
     }
+    if (failure?.classification) {
+      body.failure_kind = failure.classification.kind;
+      body.failure_provider_id = failure.classification.providerId;
+      body.failure_model_id = failure.classification.modelId;
+      body.failure_reason = failure.classification.reason;
+    }
+    if (abandonedWithoutAnswer) body.abandoned_without_answer = true;
     try {
       await this.callWithRetry(
         "marking message as failed",
@@ -12532,27 +12692,28 @@ var ChannelDriver = class _ChannelDriver {
     this.clearSubagentInvocationCaches(messageId);
   }
   /**
-   * Classify an errored turn as a model-auth failure (#736 P1-4), for `markFailed`.
+   * Classify an errored turn and extract its bounded error shape for `markFailed`.
    *
    * `messageFailure` alone (structured OpenCode error → `model_auth`) covers
    * most cases; when it returns `null` on this ALREADY-FAILED turn, fall back
    * to the P1-2b zero-provider check — one extra loopback call to
    * `hasAnyConfiguredProvider`, only reached when the structured classifier
    * couldn't place it. Fails open (never throws): a fallback probe failure
-   * (`null`/indeterminate) leaves the classification `null`, which produces
-   * today's byte-identical PATCH body via `markFailed`'s `if (failure)` guard.
+   * (`null`/indeterminate) leaves the classification `null`; the raw error shape
+   * is still retained for diagnostics.
    */
   async classifyModelAuthFailure(messages, userMessageId) {
-    const classified = messageFailure(messages, userMessageId);
-    if (classified != null) return classified;
     const reply = findLastAssistantReplyFor(messages, userMessageId);
-    const hasProvider = await this.hasAnyConfiguredProvider();
-    return applyZeroProviderFallback(
+    const errorShape = extractReplyErrorShape(reply);
+    const classified = messageFailure(messages, userMessageId);
+    if (classified != null) return { classification: classified, errorShape };
+    const classification = applyZeroProviderFallback(
       classified,
-      hasProvider,
+      await this.hasAnyConfiguredProvider(),
       reply?.info?.providerID ?? null,
       reply?.info?.modelID ?? null
     );
+    return { classification, errorShape };
   }
   async recordSubagentModelAuthFailure(failure, conversationId, messageId) {
     const providerId = failure.providerId ?? "(unknown)";
@@ -13649,6 +13810,8 @@ function createCredentialSync({
 var MAX_ACTIVITY_LOG_ENTRIES = 10;
 var MAX_DISCONNECT_REASON_LENGTH = 200;
 var CHANNEL_POLL_INTERVAL_MS = Number(process.env.EVIDENT_CHANNEL_POLL_INTERVAL_MS) || 2e3;
+var CREDENTIAL_ADOPTION_PROBE_ATTEMPTS = 3;
+var CREDENTIAL_ADOPTION_PROBE_INTERVAL_MS = 1e3;
 var CHANNEL_STUCK_QUEUED_MS = Number(process.env.EVIDENT_STUCK_QUEUED_MS) || void 0;
 var SHUTDOWN_DRAIN_TIMEOUT_MS = Number(process.env.EVIDENT_SHUTDOWN_DRAIN_MS) || 25e3;
 var TELEMETRY_SHUTDOWN_TIMEOUT_MS = 5e3;
@@ -13996,6 +14159,125 @@ async function handleAuthError(state, error2) {
     return { success: false };
   }
 }
+function logCredentialAdoptionActivity(state, activity, level) {
+  const message = (() => {
+    switch (activity.outcome) {
+      case "deferred":
+        return `CREDENTIAL-ADOPTION: deferred generation ${activity.generation} with ${activity.active_sessions} active session(s)`;
+      case "reloaded_ready":
+        return `CREDENTIAL-ADOPTION: adopted generation ${activity.generation} with ${activity.active_sessions} active session(s); readiness probe succeeded after ${activity.probe_attempts} attempt(s)`;
+      case "reloaded_interrupted": {
+        const response = activity.probe_status === null ? "without an HTTP response" : `with HTTP ${activity.probe_status} and ${activity.probe_body_empty ? "an empty" : "a non-empty"} body`;
+        return `CREDENTIAL-ADOPTION: generation ${activity.generation} with ${activity.active_sessions} active session(s); replacement instance answered ${response} after ${activity.probe_attempts} probe attempt(s)`;
+      }
+      case "reload_failed":
+        return `CREDENTIAL-ADOPTION: failed to adopt generation ${activity.generation} with ${activity.active_sessions} active session(s)`;
+      case "skipped_v2":
+        return `CREDENTIAL-ADOPTION: skipped generation ${activity.generation} with ${activity.active_sessions} active session(s) on OpenCode V2`;
+    }
+  })();
+  logActivity(state, {
+    type: level === "error" ? "error" : "info",
+    level,
+    ...level === "error" ? { error: message } : { message },
+    metadata: {
+      ...activity,
+      opencode_version: state.opencodeVersion,
+      port: state.port
+    }
+  });
+}
+async function adoptCredentialGeneration(state, generation, adoption, activeSessions, deferredForMs) {
+  const reload = await disposeAndReloadProviderConfig(
+    state.port,
+    state.opencodeClient ?? void 0
+  );
+  if (reload.outcome === "failed") {
+    logCredentialAdoptionActivity(
+      state,
+      {
+        outcome: "reload_failed",
+        generation,
+        active_sessions: activeSessions,
+        adoption,
+        deferred_for_ms: deferredForMs,
+        reload_status: reload.status,
+        reload_detail: reload.detail,
+        probe_attempts: null,
+        probe_status: null,
+        probe_body_empty: null,
+        probe_detail: null
+      },
+      "error"
+    );
+    return;
+  }
+  if (reload.outcome === "skipped_v2") {
+    logCredentialAdoptionActivity(
+      state,
+      {
+        outcome: "skipped_v2",
+        generation,
+        active_sessions: activeSessions,
+        adoption,
+        deferred_for_ms: deferredForMs,
+        reload_status: null,
+        reload_detail: null,
+        probe_attempts: null,
+        probe_status: null,
+        probe_body_empty: null,
+        probe_detail: null
+      },
+      "debug"
+    );
+    return;
+  }
+  let lastProbe = null;
+  let attempts = 0;
+  for (attempts = 1; attempts <= CREDENTIAL_ADOPTION_PROBE_ATTEMPTS; attempts++) {
+    lastProbe = await probeOpenCodeRequestReadiness(state.port, state.opencodeClient ?? void 0);
+    if (lastProbe.ok) {
+      logCredentialAdoptionActivity(
+        state,
+        {
+          outcome: "reloaded_ready",
+          generation,
+          active_sessions: activeSessions,
+          adoption,
+          deferred_for_ms: deferredForMs,
+          reload_status: null,
+          reload_detail: null,
+          probe_attempts: attempts,
+          probe_status: lastProbe.status,
+          probe_body_empty: lastProbe.bodyEmpty,
+          probe_detail: lastProbe.detail
+        },
+        "warn"
+      );
+      return;
+    }
+    if (attempts < CREDENTIAL_ADOPTION_PROBE_ATTEMPTS) {
+      await new Promise((resolve5) => setTimeout(resolve5, CREDENTIAL_ADOPTION_PROBE_INTERVAL_MS));
+    }
+  }
+  logCredentialAdoptionActivity(
+    state,
+    {
+      outcome: "reloaded_interrupted",
+      generation,
+      active_sessions: activeSessions,
+      adoption,
+      deferred_for_ms: deferredForMs,
+      reload_status: null,
+      reload_detail: null,
+      probe_attempts: CREDENTIAL_ADOPTION_PROBE_ATTEMPTS,
+      probe_status: lastProbe?.status ?? null,
+      probe_body_empty: lastProbe?.bodyEmpty ?? null,
+      probe_detail: lastProbe?.detail ?? null
+    },
+    "error"
+  );
+}
 async function driveChannels(state, driver) {
   let idlePolls = 0;
   let idleMs = 0;
@@ -14005,6 +14287,15 @@ async function driveChannels(state, driver) {
   let lastSeenAppliedFiles = driver.fileSyncActivity().appliedFiles;
   let lastSeenClaudeApplies = driver.fileSyncActivity().claudeCredentialApplies;
   let lastSeenOpencodeAuthApplies = driver.fileSyncActivity().opencodeAuthApplyCount;
+  const credentialAdoption = {
+    generation: 0,
+    pendingGeneration: null,
+    pendingSinceMs: null,
+    deferredGeneration: null,
+    adoption: 0
+  };
+  const stayAwakeHolds = () => state.stayAwakeUntilMs !== null && Date.now() < state.stayAwakeUntilMs;
+  let stayAwakeSuppressionLogged = false;
   while (state.running) {
     const cycleStartedAtMs = performance.now();
     let idleThisCycle = false;
@@ -14046,12 +14337,30 @@ async function driveChannels(state, driver) {
       lastSeenOpencodeAuthApplies = opencodeAuthApplies;
       if (opencodeAuthApplied) state.openaiUsageRearm?.();
       if (claudeCredentialApplied || opencodeAuthApplied) {
-        void reloadProviderCache(state.port, state.opencodeClient ?? void 0).catch(
-          (error2) => logActivity(state, {
-            type: "error",
-            error: `OpenCode provider cache reload failed on port ${state.port}: ${error2 instanceof Error ? error2.message : String(error2)}`
-          })
-        );
+        credentialAdoption.generation++;
+        if (state.opencodeApiVersion === "v2") {
+          logCredentialAdoptionActivity(
+            state,
+            {
+              outcome: "skipped_v2",
+              generation: credentialAdoption.generation,
+              active_sessions: driver.protectedSessionIds().size,
+              adoption: null,
+              deferred_for_ms: null,
+              reload_status: null,
+              reload_detail: null,
+              probe_attempts: null,
+              probe_status: null,
+              probe_body_empty: null,
+              probe_detail: null
+            },
+            "debug"
+          );
+        } else {
+          credentialAdoption.pendingGeneration = credentialAdoption.generation;
+          credentialAdoption.pendingSinceMs = Date.now();
+          credentialAdoption.deferredGeneration = null;
+        }
       }
       if (opencodeAuthApplied && state.opencodeApiVersion === "v2") {
         void importOpenCodeV2CredentialsForState(state).catch(
@@ -14106,6 +14415,49 @@ async function driveChannels(state, driver) {
         }
       }
     }
+    if (credentialAdoption.pendingGeneration !== null) {
+      const activeSessions = driver.protectedSessionIds();
+      const generation = credentialAdoption.pendingGeneration;
+      if (activeSessions.size > 0) {
+        if (credentialAdoption.deferredGeneration !== generation) {
+          credentialAdoption.deferredGeneration = generation;
+          logCredentialAdoptionActivity(
+            state,
+            {
+              outcome: "deferred",
+              generation,
+              active_sessions: activeSessions.size,
+              adoption: null,
+              deferred_for_ms: null,
+              reload_status: null,
+              reload_detail: null,
+              probe_attempts: null,
+              probe_status: null,
+              probe_body_empty: null,
+              probe_detail: null
+            },
+            "warn"
+          );
+        }
+      } else {
+        const deferredForMs = credentialAdoption.deferredGeneration === generation && credentialAdoption.pendingSinceMs !== null ? Math.max(0, Date.now() - credentialAdoption.pendingSinceMs) : null;
+        credentialAdoption.pendingGeneration = null;
+        credentialAdoption.pendingSinceMs = null;
+        credentialAdoption.adoption++;
+        void adoptCredentialGeneration(
+          state,
+          generation,
+          credentialAdoption.adoption,
+          activeSessions.size,
+          deferredForMs
+        ).catch(
+          (error2) => logActivity(state, {
+            type: "error",
+            error: `CREDENTIAL-ADOPTION: adoption failed for generation ${generation} with ${activeSessions.size} active session(s) on port ${state.port}: ${error2 instanceof Error ? error2.message : String(error2)}`
+          })
+        );
+      }
+    }
     await new Promise((resolve5) => setTimeout(resolve5, CHANNEL_POLL_INTERVAL_MS));
     const cycleMs = performance.now() - cycleStartedAtMs;
     if (idleThisCycle) idleMs += cycleMs;
@@ -14119,11 +14471,20 @@ async function driveChannels(state, driver) {
       if (state.interactive) displayStatus(state);
       break;
     }
-    if (state.idleTimeout !== null && idlePolls >= 2 && idleMs > state.idleTimeout * 1e3) {
+    if (state.idleTimeout !== null && idlePolls >= 2 && idleMs > state.idleTimeout * 1e3 && !stayAwakeHolds()) {
       logActivity(state, { type: "info", message: "Idle timeout reached" });
       if (state.interactive) displayStatus(state);
       break;
     }
+    if (state.idleTimeout !== null && idlePolls >= 2 && idleMs > state.idleTimeout * 1e3 && stayAwakeHolds()) {
+      if (!stayAwakeSuppressionLogged) {
+        stayAwakeSuppressionLogged = true;
+        logActivity(state, {
+          type: "info",
+          message: "Idle, but staying awake until the requested minimum passes"
+        });
+      }
+    }
   }
 }
 var SESSION_CLEANUP_FIRST_SWEEP_MS = 1e4;
@@ -14452,10 +14813,12 @@ function scheduleUsageReporting(state, params) {
   let phase = "dormant";
   let rearmRequested = false;
   const armProbe = () => {
+    if (!state.running) return;
     phase = "probe-pending";
     params.setTimer(setTimeout(() => void tick(true), params.firstDelayMs()));
   };
   const scheduleNextTick = () => {
+    if (!state.running) return;
     if (rearmRequested) {
       rearmRequested = false;
       armProbe();
@@ -14489,6 +14852,7 @@ function scheduleUsageReporting(state, params) {
     phase = "tick-in-flight";
     try {
       const usage = await params.fetchUsage();
+      if (!state.running) return;
       const result = await params.report(usage);
       if (result.ok) {
         if (consecutiveFailures > 0) {
@@ -14627,6 +14991,7 @@ function scheduleResourceUsageReporting(state, options) {
           message: lowDiskDecision.warning
         });
       }
+      if (!state.running) return;
       const result = await reportResourceUsage(state.agentId, state.authHeader, usage);
       if (result.ok) {
         if (consecutiveFailures > 0) {
@@ -14659,7 +15024,9 @@ function scheduleResourceUsageReporting(state, options) {
         message: `Resource usage reporting failed: ${message}${failureStreakSuffix(consecutiveFailures)}`
       });
     } finally {
-      state.resourceUsageTimer = setTimeout(() => void tick(), usageReportDelayMs());
+      if (state.running) {
+        state.resourceUsageTimer = setTimeout(() => void tick(), usageReportDelayMs());
+      }
     }
   };
   state.resourceUsageTimer = setTimeout(() => void tick(), firstReportDelayMs());
@@ -14862,6 +15229,8 @@ async function run(options) {
     port: options.port ?? 4096,
     conversationFilter: options.conversation ?? null,
     idleTimeout: options.idleTimeout ?? null,
+    stayAwakeSeconds: null,
+    stayAwakeUntilMs: null,
     json: options.json ?? false,
     interactive,
     logLevel,
@@ -15051,6 +15420,17 @@ async function run(options) {
     }
     spinner?.succeed(`Runner: ${validation.agent.name || state.agentId}`);
     state.agentName = validation.agent.name;
+    const requestedStayAwakeSeconds = validation.agent?.wake_stay_awake_seconds;
+    if (typeof requestedStayAwakeSeconds === "number" && Number.isFinite(requestedStayAwakeSeconds) && requestedStayAwakeSeconds > 0) {
+      state.stayAwakeSeconds = requestedStayAwakeSeconds;
+    } else if (requestedStayAwakeSeconds !== void 0 && requestedStayAwakeSeconds !== null) {
+      state.stayAwakeSeconds = null;
+      log2(
+        state,
+        `Ignoring invalid wake_stay_awake_seconds (type: ${typeof requestedStayAwakeSeconds})`,
+        "warn"
+      );
+    }
     const microvmId = process.env.MICROVM_ID?.trim();
     if (microvmId) {
       const reported = await reportMicrovmId(state.agentId, state.authHeader, microvmId);
@@ -15294,7 +15674,7 @@ async function run(options) {
             logActivity(state, { type: "info", level: "warn", message: versionWarning });
           }
         }
-        await reloadProviderCache(state.port, state.opencodeClient ?? void 0);
+        await disposeAndReloadProviderConfig(state.port, state.opencodeClient ?? void 0);
         if (state.opencodeApiVersion === "v2" && state.opencodeClient !== null) {
           await importOpenCodeV2CredentialsForState(state);
         }
@@ -15465,6 +15845,13 @@ async function run(options) {
             type: "info",
             message: connectionMessage
           });
+          if (!isReconnect && state.stayAwakeSeconds !== null && state.stayAwakeUntilMs === null) {
+            state.stayAwakeUntilMs = Date.now() + state.stayAwakeSeconds * 1e3;
+            logActivity(state, {
+              type: "info",
+              message: `Staying awake for at least ${state.stayAwakeSeconds}s as requested by the wake`
+            });
+          }
           if (options.tunnelReadyFile) {
             const marker = writeTunnelReadyMarker(options.tunnelReadyFile, agentId);
             if (marker.ok) {
@@ -15599,6 +15986,7 @@ async function run(options) {
       if (error2.message === "Unauthorized") tunnelSpinner?.fail("Unauthorized");
       throw error2;
     }
+    if (!state.running) return;
     scheduleSessionCleanup(state, channelDriver, options);
     state.claudeUsageRearm = scheduleClaudeUsageReporting(state, options);
     state.openaiUsageRearm = scheduleUsageReporting(state, {