@poncho-ai/harness 0.28.1 → 0.28.2

package/.turbo/turbo-build.log

@@ -1,5 +1,5 @@
 
-> @poncho-ai/harness@0.28.1 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 288.74 KB
-ESM ⚡️ Build success in 135ms
+ESM dist/index.js 289.62 KB
+ESM ⚡️ Build success in 213ms
 DTS Build start
-DTS ⚡️ Build success in 6964ms
+DTS ⚡️ Build success in 7196ms
 DTS dist/index.d.ts 29.62 KB

package/CHANGELOG.md

@@ -1,5 +1,17 @@
 # @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

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 |
@@ -6347,6 +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({

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@poncho-ai/harness",
-  "version": "0.28.1",
+  "version": "0.28.2",
   "description": "Agent execution runtime - conversation loop, tool dispatch, streaming",
   "repository": {
     "type": "git",

package/src/harness.ts

@@ -2030,6 +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;