@poncho-ai/harness 0.28.0 → 0.28.2

package/.turbo/turbo-build.log

@@ -1,5 +1,5 @@
 
-> @poncho-ai/harness@0.28.0 build /home/runner/work/poncho-ai/poncho-ai/packages/harness
+> @poncho-ai/harness@0.28.2 build /home/runner/work/poncho-ai/poncho-ai/packages/harness
 > node scripts/embed-docs.js && tsup src/index.ts --format esm --dts
 
 [embed-docs] Generated poncho-docs.ts with 4 topics
@@ -8,8 +8,8 @@
 CLI tsup v8.5.1
 CLI Target: es2022
 ESM Build start
-ESM dist/index.js 284.05 KB
-ESM ⚡️ Build success in 147ms
+ESM dist/index.js 289.62 KB
+ESM ⚡️ Build success in 213ms
 DTS Build start
-DTS ⚡️ Build success in 7926ms
-DTS dist/index.d.ts 29.26 KB
+DTS ⚡️ Build success in 7196ms
+DTS dist/index.d.ts 29.62 KB

package/CHANGELOG.md

@@ -1,5 +1,28 @@
 # @poncho-ai/harness
 
+## 0.28.2
+
+### Patch Changes
+
+- [`98df42f`](https://github.com/cesr/poncho-ai/commit/98df42f79e0a376d0a864598557758bfa644039d) Thanks [@cesr](https://github.com/cesr)! - Fix serverless subagent and continuation reliability
+  - Use stable internal secret across serverless instances for callback auth
+  - Wrap continuation self-fetches in waitUntil to survive function shutdown
+  - Set runStatus during callback re-runs so clients detect active processing
+  - Add post-streaming soft deadline check to catch long model responses
+  - Client auto-recovers from abrupt stream termination and orphaned continuations
+  - Fix callback continuation losing \_continuationMessages when no pending results
+
+## 0.28.1
+
+### Patch Changes
+
+- [`4d50ad9`](https://github.com/cesr/poncho-ai/commit/4d50ad970886c9d3635ec36a407514c91ce6a71a) Thanks [@cesr](https://github.com/cesr)! - Improve callback-run reliability and streaming across subagent workflows, including safer concurrent approval handling and parent callback retriggers.
+
+  Add context window/token reporting through run completion events, improve cron/web UI rendering and approval streaming behavior, and harden built-in web search retry/throttle behavior.
+
+- Updated dependencies [[`4d50ad9`](https://github.com/cesr/poncho-ai/commit/4d50ad970886c9d3635ec36a407514c91ce6a71a)]:
+  - @poncho-ai/sdk@1.6.1
+
 ## 0.28.0
 
 ### Minor Changes

package/dist/index.d.ts

@@ -188,6 +188,12 @@ interface ConversationStore {
     rename(conversationId: string, title: string): Promise<Conversation | undefined>;
     delete(conversationId: string): Promise<boolean>;
     appendSubagentResult(conversationId: string, result: PendingSubagentResult): Promise<void>;
+    /**
+     * Atomically clear `runningCallbackSince` without clobbering other fields.
+     * Returns the conversation as it exists after the clear (with current
+     * `pendingSubagentResults`).
+     */
+    clearCallbackLock(conversationId: string): Promise<Conversation | undefined>;
 }
 type StateProviderName = "local" | "memory" | "redis" | "upstash" | "dynamodb";
 interface StateConfig {
@@ -221,6 +227,7 @@ declare class InMemoryConversationStore implements ConversationStore {
     rename(conversationId: string, title: string): Promise<Conversation | undefined>;
     delete(conversationId: string): Promise<boolean>;
     appendSubagentResult(conversationId: string, result: PendingSubagentResult): Promise<void>;
+    clearCallbackLock(conversationId: string): Promise<Conversation | undefined>;
 }
 type ConversationSummary = {
     conversationId: string;

package/dist/index.js

@@ -1604,6 +1604,8 @@ Remote storage keys are namespaced and versioned, for example \`poncho:v1:<agent
 | \`ANTHROPIC_API_KEY\` | Yes* | Claude API key |
 | \`OPENAI_API_KEY\` | No | OpenAI API key (if using OpenAI) |
 | \`PONCHO_AUTH_TOKEN\` | No | Unified auth token (Web UI passphrase + API Bearer token) |
+| \`PONCHO_INTERNAL_SECRET\` | No | Shared secret used by internal serverless callbacks (recommended for Vercel/Lambda) |
+| \`PONCHO_SELF_BASE_URL\` | No | Explicit base URL for internal self-callbacks when auto-detection is unavailable |
 | \`OTEL_EXPORTER_OTLP_ENDPOINT\` | No | Telemetry destination |
 | \`LATITUDE_API_KEY\` | No | Latitude dashboard integration |
 | \`LATITUDE_PROJECT_ID\` | No | Latitude project identifier for capture traces |
@@ -4211,15 +4213,16 @@ var loadRunnableScriptFunction = async (scriptPath) => {
 };
 var loadScriptModule = async (scriptPath) => {
   const extension = extname(scriptPath).toLowerCase();
+  const cacheBust = `?t=${Date.now()}`;
   if (extension === ".ts" || extension === ".mts" || extension === ".cts") {
-    const jiti = createJiti2(import.meta.url, { interopDefault: true });
-    return await jiti.import(scriptPath);
+    const jiti = createJiti2(import.meta.url, { interopDefault: true, moduleCache: false });
+    return await jiti.import(scriptPath + cacheBust);
   }
   try {
-    return await import(pathToFileURL(scriptPath).href);
+    return await import(pathToFileURL(scriptPath).href + cacheBust);
   } catch {
-    const jiti = createJiti2(import.meta.url, { interopDefault: true });
-    return await jiti.import(scriptPath);
+    const jiti = createJiti2(import.meta.url, { interopDefault: true, moduleCache: false });
+    return await jiti.import(scriptPath + cacheBust);
   }
 };
 var extractRunnableFunction = (value) => {
@@ -4263,21 +4266,93 @@ import { load as cheerioLoad } from "cheerio";
 import { defineTool as defineTool5 } from "@poncho-ai/sdk";
 var SEARCH_UA = "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/124.0.0.0 Safari/537.36";
 var FETCH_TIMEOUT_MS = 15e3;
+var SEARCH_MAX_RETRIES = 4;
+var SEARCH_INITIAL_DELAY_MS = 2e3;
+var SEARCH_MIN_INTERVAL_MS = 4e3;
+var SEARCH_FALLBACK_COOLDOWN_MS = 12e3;
+var sleep = (ms) => new Promise((r) => setTimeout(r, ms));
+var searchQueue = Promise.resolve();
+var nextSearchAllowedAt = 0;
+function parseRetryAfterMs(retryAfterHeader) {
+  if (!retryAfterHeader) return SEARCH_FALLBACK_COOLDOWN_MS;
+  const asSeconds = Number(retryAfterHeader);
+  if (Number.isFinite(asSeconds) && asSeconds >= 0) {
+    return Math.max(Math.floor(asSeconds * 1e3), SEARCH_MIN_INTERVAL_MS);
+  }
+  const asDate = new Date(retryAfterHeader).getTime();
+  if (Number.isFinite(asDate)) {
+    return Math.max(asDate - Date.now(), SEARCH_MIN_INTERVAL_MS);
+  }
+  return SEARCH_FALLBACK_COOLDOWN_MS;
+}
+function applyRateLimitCooldown(retryAfterHeader) {
+  const cooldownMs = parseRetryAfterMs(retryAfterHeader);
+  nextSearchAllowedAt = Math.max(nextSearchAllowedAt, Date.now() + cooldownMs);
+}
+async function runWithSearchThrottle(fn) {
+  const previous = searchQueue;
+  let release;
+  searchQueue = new Promise((resolve12) => {
+    release = resolve12;
+  });
+  await previous.catch(() => {
+  });
+  try {
+    const waitMs = nextSearchAllowedAt - Date.now();
+    if (waitMs > 0) {
+      await sleep(waitMs);
+    }
+    const result = await fn();
+    nextSearchAllowedAt = Math.max(nextSearchAllowedAt, Date.now() + SEARCH_MIN_INTERVAL_MS);
+    return result;
+  } finally {
+    release?.();
+  }
+}
+function isRetryableStatus(status) {
+  return status === 429 || status === 503 || status >= 500;
+}
 async function braveSearch(query, maxResults) {
   const url = `https://search.brave.com/search?q=${encodeURIComponent(query)}`;
-  const res = await fetch(url, {
-    headers: {
-      "User-Agent": SEARCH_UA,
-      Accept: "text/html,application/xhtml+xml",
-      "Accept-Language": "en-US,en;q=0.9"
-    },
-    signal: AbortSignal.timeout(FETCH_TIMEOUT_MS)
-  });
-  if (!res.ok) {
-    throw new Error(`Search request failed (${res.status} ${res.statusText})`);
+  let lastError;
+  for (let attempt = 0; attempt < SEARCH_MAX_RETRIES; attempt++) {
+    if (attempt > 0) {
+      const delay = SEARCH_INITIAL_DELAY_MS * 2 ** (attempt - 1) + Math.floor(Math.random() * 500);
+      await sleep(delay);
+    }
+    try {
+      const html = await runWithSearchThrottle(async () => {
+        const res = await fetch(url, {
+          headers: {
+            "User-Agent": SEARCH_UA,
+            Accept: "text/html,application/xhtml+xml",
+            "Accept-Language": "en-US,en;q=0.9"
+          },
+          signal: AbortSignal.timeout(FETCH_TIMEOUT_MS)
+        });
+        if (!res.ok) {
+          if (res.status === 429) {
+            applyRateLimitCooldown(res.headers.get("retry-after"));
+          }
+          const error = new Error(`Search request failed (${res.status} ${res.statusText})`);
+          if (isRetryableStatus(res.status)) {
+            throw error;
+          }
+          error.retryable = false;
+          throw error;
+        }
+        return await res.text();
+      });
+      return parseBraveResults(html, maxResults);
+    } catch (err) {
+      lastError = err instanceof Error ? err : new Error(String(err));
+      if (lastError.retryable === false) {
+        throw lastError;
+      }
+      if (attempt < SEARCH_MAX_RETRIES - 1) continue;
+    }
   }
-  const html = await res.text();
-  return parseBraveResults(html, maxResults);
+  throw lastError ?? new Error("Search failed after retries");
 }
 function parseBraveResults(html, max) {
   const $ = cheerioLoad(html);
@@ -5878,6 +5953,8 @@ ${this.skillFingerprint}`;
     let totalOutputTokens = 0;
     let totalCachedTokens = 0;
     let transientStepRetryCount = 0;
+    let latestContextTokens = 0;
+    let toolOutputEstimateSinceModel = 0;
     let cachedCoreMessages = [];
     let convertedUpTo = 0;
     for (let step = 1; step <= maxSteps; step += 1) {
@@ -5907,7 +5984,9 @@ ${this.skillFingerprint}`;
             duration: now() - start,
             continuation: true,
             continuationMessages: [...messages],
-            maxSteps
+            maxSteps,
+            contextTokens: latestContextTokens + toolOutputEstimateSinceModel,
+            contextWindow
           };
           yield pushEvent({ type: "run:completed", runId, result: result2 });
           return;
@@ -6147,10 +6226,13 @@ ${textContent}` };
               if (emittedMessages.length > 0 && emittedMessages[emittedMessages.length - 1].role === "user") {
                 emittedMessages.pop();
               }
+              const tokensAfterCompaction = estimateTotalTokens(integrityPrompt, messages, toolDefsJson);
+              latestContextTokens = tokensAfterCompaction;
+              toolOutputEstimateSinceModel = 0;
               yield pushEvent({
                 type: "compaction:completed",
                 tokensBefore: effectiveTokens,
-                tokensAfter: estimateTotalTokens(integrityPrompt, messages, toolDefsJson),
+                tokensAfter: tokensAfterCompaction,
                 messagesBefore: compactResult.messagesBefore,
                 compactedMessages: emittedMessages,
                 messagesAfter: compactResult.messagesAfter
@@ -6267,6 +6349,22 @@ ${textContent}` };
           yield emitCancellation();
           return;
         }
+        if (softDeadlineMs > 0 && now() - start > softDeadlineMs) {
+          const result_ = {
+            status: "completed",
+            response: responseText + fullText,
+            steps: step,
+            tokens: { input: totalInputTokens, output: totalOutputTokens, cached: totalCachedTokens },
+            duration: now() - start,
+            continuation: true,
+            continuationMessages: [...messages],
+            maxSteps,
+            contextTokens: latestContextTokens + toolOutputEstimateSinceModel,
+            contextWindow
+          };
+          yield pushEvent({ type: "run:completed", runId, result: result_ });
+          return;
+        }
         const finishReason = await result.finishReason;
         if (finishReason === "error") {
           yield pushEvent({
@@ -6297,13 +6395,16 @@ ${textContent}` };
         const usage = await result.usage;
         const toolCallsResult = await result.toolCalls;
         const stepCachedTokens = usage.inputTokenDetails?.cacheReadTokens ?? 0;
-        totalInputTokens += usage.inputTokens ?? 0;
+        const stepInputTokens = usage.inputTokens ?? 0;
+        totalInputTokens += stepInputTokens;
         totalOutputTokens += usage.outputTokens ?? 0;
         totalCachedTokens += stepCachedTokens;
+        latestContextTokens = stepInputTokens;
+        toolOutputEstimateSinceModel = 0;
         yield pushEvent({
           type: "model:response",
           usage: {
-            input: usage.inputTokens ?? 0,
+            input: stepInputTokens,
             output: usage.outputTokens ?? 0,
             cached: stepCachedTokens
           }
@@ -6349,7 +6450,9 @@ ${textContent}` };
               output: totalOutputTokens,
               cached: totalCachedTokens
             },
-            duration: now() - start
+            duration: now() - start,
+            contextTokens: latestContextTokens + toolOutputEstimateSinceModel,
+            contextWindow
           };
           yield pushEvent({ type: "run:completed", runId, result: result2 });
           return;
@@ -6477,6 +6580,7 @@ ${textContent}` };
             span?.end({ result: { value: result2.output ?? null, isError: false } });
             const serialized = JSON.stringify(result2.output ?? null);
             const outputTokenEstimate = Math.ceil(serialized.length / 4);
+            toolOutputEstimateSinceModel += outputTokenEstimate;
             yield pushEvent({
               type: "tool:completed",
               tool: result2.tool,
@@ -6587,7 +6691,9 @@ ${this.skillFingerprint}`;
         duration: now() - start,
         continuation: true,
         continuationMessages: [...messages],
-        maxSteps
+        maxSteps,
+        contextTokens: latestContextTokens + toolOutputEstimateSinceModel,
+        contextWindow
       };
       yield pushEvent({ type: "run:completed", runId, result });
     } else {
@@ -6915,6 +7021,13 @@ var InMemoryConversationStore = class {
     conversation.pendingSubagentResults.push(result);
     conversation.updatedAt = Date.now();
   }
+  async clearCallbackLock(conversationId) {
+    const conversation = this.conversations.get(conversationId);
+    if (!conversation) return void 0;
+    conversation.runningCallbackSince = void 0;
+    conversation.updatedAt = Date.now();
+    return conversation;
+  }
 };
 var FileConversationStore = class {
   workingDir;
@@ -7137,6 +7250,28 @@ var FileConversationStore = class {
     conversation.updatedAt = Date.now();
     await this.update(conversation);
   }
+  async clearCallbackLock(conversationId) {
+    await this.ensureLoaded();
+    const summary = this.conversations.get(conversationId);
+    if (!summary) return void 0;
+    const { conversationsDir } = await this.resolvePaths();
+    const filePath = resolve11(conversationsDir, summary.fileName);
+    let result;
+    this.writing = this.writing.then(async () => {
+      const conv = await this.readConversationFile(summary.fileName);
+      if (!conv) return;
+      conv.runningCallbackSince = void 0;
+      conv.updatedAt = Date.now();
+      await writeJsonAtomic3(filePath, conv);
+      this.conversations.set(conversationId, {
+        ...summary,
+        updatedAt: conv.updatedAt
+      });
+      result = conv;
+    });
+    await this.writing;
+    return result;
+  }
 };
 var FileStateStore = class {
   workingDir;
@@ -7477,6 +7612,18 @@ var KeyValueConversationStoreBase = class {
       await this.update(conversation);
     });
   }
+  async clearCallbackLock(conversationId) {
+    let result;
+    await this.withAppendLock(conversationId, async () => {
+      const conversation = await this.get(conversationId);
+      if (!conversation) return;
+      conversation.runningCallbackSince = void 0;
+      conversation.updatedAt = Date.now();
+      await this.update(conversation);
+      result = conversation;
+    });
+    return result;
+  }
 };
 var UpstashConversationStore = class extends KeyValueConversationStoreBase {
   baseUrl;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@poncho-ai/harness",
-  "version": "0.28.0",
+  "version": "0.28.2",
   "description": "Agent execution runtime - conversation loop, tool dispatch, streaming",
   "repository": {
     "type": "git",
@@ -32,7 +32,7 @@
     "redis": "^5.10.0",
     "yaml": "^2.4.0",
     "zod": "^3.22.0",
-    "@poncho-ai/sdk": "1.6.0"
+    "@poncho-ai/sdk": "1.6.1"
   },
   "devDependencies": {
     "@types/mustache": "^4.2.6",

package/src/harness.ts

@@ -1554,6 +1554,8 @@ ${boundedMainMemory.trim()}`
     let totalOutputTokens = 0;
     let totalCachedTokens = 0;
     let transientStepRetryCount = 0;
+    let latestContextTokens = 0;
+    let toolOutputEstimateSinceModel = 0;
     let cachedCoreMessages: ModelMessage[] = [];
     let convertedUpTo = 0;
 
@@ -1585,6 +1587,8 @@ ${boundedMainMemory.trim()}`
             continuation: true,
             continuationMessages: [...messages],
             maxSteps,
+            contextTokens: latestContextTokens + toolOutputEstimateSinceModel,
+            contextWindow,
           };
           yield pushEvent({ type: "run:completed", runId, result });
           return;
@@ -1885,10 +1889,13 @@ ${boundedMainMemory.trim()}`
               if (emittedMessages.length > 0 && emittedMessages[emittedMessages.length - 1].role === "user") {
                 emittedMessages.pop();
               }
+              const tokensAfterCompaction = estimateTotalTokens(integrityPrompt, messages, toolDefsJson);
+              latestContextTokens = tokensAfterCompaction;
+              toolOutputEstimateSinceModel = 0;
               yield pushEvent({
                 type: "compaction:completed",
                 tokensBefore: effectiveTokens,
-                tokensAfter: estimateTotalTokens(integrityPrompt, messages, toolDefsJson),
+                tokensAfter: tokensAfterCompaction,
                 messagesBefore: compactResult.messagesBefore!,
                 compactedMessages: emittedMessages,
                 messagesAfter: compactResult.messagesAfter!,
@@ -2023,6 +2030,25 @@ ${boundedMainMemory.trim()}`
           return;
         }
 
+      // Post-streaming soft deadline: if the model stream took long enough to
+      // push past the soft deadline, checkpoint now before tool execution.
+      if (softDeadlineMs > 0 && now() - start > softDeadlineMs) {
+        const result_: RunResult = {
+          status: "completed",
+          response: responseText + fullText,
+          steps: step,
+          tokens: { input: totalInputTokens, output: totalOutputTokens, cached: totalCachedTokens },
+          duration: now() - start,
+          continuation: true,
+          continuationMessages: [...messages],
+          maxSteps,
+          contextTokens: latestContextTokens + toolOutputEstimateSinceModel,
+          contextWindow,
+        };
+        yield pushEvent({ type: "run:completed", runId, result: result_ });
+        return;
+      }
+
       // Check finish reason for error / abnormal completions.
       const finishReason = await result.finishReason;
 
@@ -2060,14 +2086,17 @@ ${boundedMainMemory.trim()}`
 
       // Update token usage
       const stepCachedTokens = usage.inputTokenDetails?.cacheReadTokens ?? 0;
-      totalInputTokens += usage.inputTokens ?? 0;
+      const stepInputTokens = usage.inputTokens ?? 0;
+      totalInputTokens += stepInputTokens;
       totalOutputTokens += usage.outputTokens ?? 0;
       totalCachedTokens += stepCachedTokens;
+      latestContextTokens = stepInputTokens;
+      toolOutputEstimateSinceModel = 0;
 
       yield pushEvent({
         type: "model:response",
         usage: {
-          input: usage.inputTokens ?? 0,
+          input: stepInputTokens,
           output: usage.outputTokens ?? 0,
           cached: stepCachedTokens,
         },
@@ -2120,6 +2149,8 @@ ${boundedMainMemory.trim()}`
             cached: totalCachedTokens,
           },
           duration: now() - start,
+          contextTokens: latestContextTokens + toolOutputEstimateSinceModel,
+          contextWindow,
         };
         yield pushEvent({ type: "run:completed", runId, result });
         return;
@@ -2290,6 +2321,7 @@ ${boundedMainMemory.trim()}`
           span?.end({ result: { value: result.output ?? null, isError: false } });
           const serialized = JSON.stringify(result.output ?? null);
           const outputTokenEstimate = Math.ceil(serialized.length / 4);
+          toolOutputEstimateSinceModel += outputTokenEstimate;
           yield pushEvent({
             type: "tool:completed",
             tool: result.tool,
@@ -2415,6 +2447,8 @@ ${boundedMainMemory.trim()}`
         continuation: true,
         continuationMessages: [...messages],
         maxSteps,
+        contextTokens: latestContextTokens + toolOutputEstimateSinceModel,
+        contextWindow,
       };
       yield pushEvent({ type: "run:completed", runId, result });
     } else {

package/src/search-tools.ts

@@ -6,6 +6,13 @@ const SEARCH_UA =
 
 const FETCH_TIMEOUT_MS = 15_000;
 
+const SEARCH_MAX_RETRIES = 4;
+const SEARCH_INITIAL_DELAY_MS = 2_000;
+const SEARCH_MIN_INTERVAL_MS = 4_000;
+const SEARCH_FALLBACK_COOLDOWN_MS = 12_000;
+
+const sleep = (ms: number) => new Promise<void>((r) => setTimeout(r, ms));
+
 // ---------------------------------------------------------------------------
 // web_search — Brave Search HTML scraping (no API key)
 // ---------------------------------------------------------------------------
@@ -16,21 +23,100 @@ interface SearchResult {
   snippet: string;
 }
 
+let searchQueue: Promise<void> = Promise.resolve();
+let nextSearchAllowedAt = 0;
+
+function parseRetryAfterMs(retryAfterHeader: string | null): number {
+  if (!retryAfterHeader) return SEARCH_FALLBACK_COOLDOWN_MS;
+  const asSeconds = Number(retryAfterHeader);
+  if (Number.isFinite(asSeconds) && asSeconds >= 0) {
+    return Math.max(Math.floor(asSeconds * 1000), SEARCH_MIN_INTERVAL_MS);
+  }
+  const asDate = new Date(retryAfterHeader).getTime();
+  if (Number.isFinite(asDate)) {
+    return Math.max(asDate - Date.now(), SEARCH_MIN_INTERVAL_MS);
+  }
+  return SEARCH_FALLBACK_COOLDOWN_MS;
+}
+
+function applyRateLimitCooldown(retryAfterHeader: string | null): void {
+  const cooldownMs = parseRetryAfterMs(retryAfterHeader);
+  nextSearchAllowedAt = Math.max(nextSearchAllowedAt, Date.now() + cooldownMs);
+}
+
+async function runWithSearchThrottle<T>(fn: () => Promise<T>): Promise<T> {
+  const previous = searchQueue;
+  let release: (() => void) | undefined;
+  searchQueue = new Promise<void>((resolve) => {
+    release = resolve;
+  });
+
+  await previous.catch(() => {});
+  try {
+    const waitMs = nextSearchAllowedAt - Date.now();
+    if (waitMs > 0) {
+      await sleep(waitMs);
+    }
+    const result = await fn();
+    nextSearchAllowedAt = Math.max(nextSearchAllowedAt, Date.now() + SEARCH_MIN_INTERVAL_MS);
+    return result;
+  } finally {
+    release?.();
+  }
+}
+
+function isRetryableStatus(status: number): boolean {
+  return status === 429 || status === 503 || status >= 500;
+}
+
 async function braveSearch(query: string, maxResults: number): Promise<SearchResult[]> {
   const url = `https://search.brave.com/search?q=${encodeURIComponent(query)}`;
-  const res = await fetch(url, {
-    headers: {
-      "User-Agent": SEARCH_UA,
-      Accept: "text/html,application/xhtml+xml",
-      "Accept-Language": "en-US,en;q=0.9",
-    },
-    signal: AbortSignal.timeout(FETCH_TIMEOUT_MS),
-  });
-  if (!res.ok) {
-    throw new Error(`Search request failed (${res.status} ${res.statusText})`);
+
+  let lastError: Error | undefined;
+  for (let attempt = 0; attempt < SEARCH_MAX_RETRIES; attempt++) {
+    if (attempt > 0) {
+      const delay = SEARCH_INITIAL_DELAY_MS * 2 ** (attempt - 1) + Math.floor(Math.random() * 500);
+      await sleep(delay);
+    }
+
+    try {
+      const html = await runWithSearchThrottle(async () => {
+        const res = await fetch(url, {
+          headers: {
+            "User-Agent": SEARCH_UA,
+            Accept: "text/html,application/xhtml+xml",
+            "Accept-Language": "en-US,en;q=0.9",
+          },
+          signal: AbortSignal.timeout(FETCH_TIMEOUT_MS),
+        });
+
+        if (!res.ok) {
+          if (res.status === 429) {
+            applyRateLimitCooldown(res.headers.get("retry-after"));
+          }
+          const error = new Error(`Search request failed (${res.status} ${res.statusText})`);
+          if (isRetryableStatus(res.status)) {
+            throw error;
+          }
+          // Non-retryable status: surface immediately.
+          (error as Error & { retryable?: boolean }).retryable = false;
+          throw error;
+        }
+
+        return await res.text();
+      });
+
+      return parseBraveResults(html, maxResults);
+    } catch (err) {
+      lastError = err instanceof Error ? err : new Error(String(err));
+      if ((lastError as Error & { retryable?: boolean }).retryable === false) {
+        throw lastError;
+      }
+      if (attempt < SEARCH_MAX_RETRIES - 1) continue;
+    }
   }
-  const html = await res.text();
-  return parseBraveResults(html, maxResults);
+
+  throw lastError ?? new Error("Search failed after retries");
 }
 
 function parseBraveResults(html: string, max: number): SearchResult[] {

package/src/skill-tools.ts

@@ -412,17 +412,19 @@ const loadRunnableScriptFunction = async (
 
 const loadScriptModule = async (scriptPath: string): Promise<unknown> => {
   const extension = extname(scriptPath).toLowerCase();
-  // Node emits noisy warnings when attempting native ESM import on TypeScript
-  // files in serverless environments. Use jiti first for TS entrypoints.
+  // Both Node's native import() and jiti cache modules by URL/path.
+  // Append a cache-busting query string so edits made by the agent are
+  // picked up on the next run_skill_script call.
+  const cacheBust = `?t=${Date.now()}`;
   if (extension === ".ts" || extension === ".mts" || extension === ".cts") {
-    const jiti = createJiti(import.meta.url, { interopDefault: true });
-    return await jiti.import(scriptPath);
+    const jiti = createJiti(import.meta.url, { interopDefault: true, moduleCache: false });
+    return await jiti.import(scriptPath + cacheBust);
   }
   try {
-    return await import(pathToFileURL(scriptPath).href);
+    return await import(pathToFileURL(scriptPath).href + cacheBust);
   } catch {
-    const jiti = createJiti(import.meta.url, { interopDefault: true });
-    return await jiti.import(scriptPath);
+    const jiti = createJiti(import.meta.url, { interopDefault: true, moduleCache: false });
+    return await jiti.import(scriptPath + cacheBust);
   }
 };
 

package/src/state.ts

@@ -84,6 +84,12 @@ export interface ConversationStore {
   rename(conversationId: string, title: string): Promise<Conversation | undefined>;
   delete(conversationId: string): Promise<boolean>;
   appendSubagentResult(conversationId: string, result: PendingSubagentResult): Promise<void>;
+  /**
+   * Atomically clear `runningCallbackSince` without clobbering other fields.
+   * Returns the conversation as it exists after the clear (with current
+   * `pendingSubagentResults`).
+   */
+  clearCallbackLock(conversationId: string): Promise<Conversation | undefined>;
 }
 
 export type StateProviderName =
@@ -325,6 +331,14 @@ export class InMemoryConversationStore implements ConversationStore {
     conversation.pendingSubagentResults.push(result);
     conversation.updatedAt = Date.now();
   }
+
+  async clearCallbackLock(conversationId: string): Promise<Conversation | undefined> {
+    const conversation = this.conversations.get(conversationId);
+    if (!conversation) return undefined;
+    conversation.runningCallbackSince = undefined;
+    conversation.updatedAt = Date.now();
+    return conversation;
+  }
 }
 
 export type ConversationSummary = {
@@ -607,6 +621,31 @@ class FileConversationStore implements ConversationStore {
     conversation.updatedAt = Date.now();
     await this.update(conversation);
   }
+
+  async clearCallbackLock(conversationId: string): Promise<Conversation | undefined> {
+    await this.ensureLoaded();
+    const summary = this.conversations.get(conversationId);
+    if (!summary) return undefined;
+    const { conversationsDir } = await this.resolvePaths();
+    const filePath = resolve(conversationsDir, summary.fileName);
+    let result: Conversation | undefined;
+    // Read inside the writing chain so we see the latest state after any
+    // pending appendSubagentResult writes have flushed.
+    this.writing = this.writing.then(async () => {
+      const conv = await this.readConversationFile(summary.fileName);
+      if (!conv) return;
+      conv.runningCallbackSince = undefined;
+      conv.updatedAt = Date.now();
+      await writeJsonAtomic(filePath, conv);
+      this.conversations.set(conversationId, {
+        ...summary,
+        updatedAt: conv.updatedAt,
+      });
+      result = conv;
+    });
+    await this.writing;
+    return result;
+  }
 }
 
 type LocalStateFile = {
@@ -1005,6 +1044,19 @@ abstract class KeyValueConversationStoreBase implements ConversationStore {
       await this.update(conversation);
     });
   }
+
+  async clearCallbackLock(conversationId: string): Promise<Conversation | undefined> {
+    let result: Conversation | undefined;
+    await this.withAppendLock(conversationId, async () => {
+      const conversation = await this.get(conversationId);
+      if (!conversation) return;
+      conversation.runningCallbackSince = undefined;
+      conversation.updatedAt = Date.now();
+      await this.update(conversation);
+      result = conversation;
+    });
+    return result;
+  }
 }
 
 class UpstashConversationStore extends KeyValueConversationStoreBase {