@poncho-ai/harness 0.29.0 → 0.30.0

package/.turbo/turbo-build.log

@@ -1,5 +1,5 @@
 
-> @poncho-ai/harness@0.29.0 build /Users/cesar/Dev/latitude/poncho-ai/packages/harness
+> @poncho-ai/harness@0.30.0 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 297.56 KB
-ESM ⚡️ Build success in 32ms
+ESM dist/index.js 300.00 KB
+ESM ⚡️ Build success in 135ms
 DTS Build start
-DTS ⚡️ Build success in 4608ms
-DTS dist/index.d.ts 30.41 KB
+DTS ⚡️ Build success in 7526ms
+DTS dist/index.d.ts 30.64 KB

package/CHANGELOG.md

@@ -1,5 +1,16 @@
 # @poncho-ai/harness
 
+## 0.30.0
+
+### Minor Changes
+
+- [`193c367`](https://github.com/cesr/poncho-ai/commit/193c367568dce22a470dff6acd022c221be3b722) Thanks [@cesr](https://github.com/cesr)! - Unified continuation logic across all entry points (chat, cron, subagents, SDK) with mid-stream soft deadline checkpointing and proper context preservation across continuation boundaries.
+
+### Patch Changes
+
+- Updated dependencies [[`193c367`](https://github.com/cesr/poncho-ai/commit/193c367568dce22a470dff6acd022c221be3b722)]:
+  - @poncho-ai/sdk@1.6.3
+
 ## 0.29.0
 
 ### Minor Changes

package/dist/index.d.ts

@@ -176,6 +176,10 @@ interface Conversation {
     /** Harness-internal message chain preserved across continuation runs.
      *  Cleared when a run completes without continuation. */
     _continuationMessages?: Message[];
+    /** Number of continuation pickups for the current multi-step run.
+     *  Reset when a run completes without continuation. Used to enforce
+     *  a maximum continuation count across all entry points. */
+    _continuationCount?: number;
     /** Full structured message chain from the last harness run, including
      *  tool-call and tool-result messages the model needs for context.
      *  Unlike `_continuationMessages`, this is always set after a run

package/dist/index.js

@@ -6131,7 +6131,7 @@ ${this.skillFingerprint}`;
       if (lastMsg && lastMsg.role !== "user") {
         messages.push({
           role: "user",
-          content: "[System: Your previous turn was interrupted by a time limit. Continue from where you left off \u2014 do NOT repeat what you already said. Proceed directly with the next action or tool call.]",
+          content: "[System: Your previous turn was interrupted by a time limit. Your partial response above is already visible to the user. Continue EXACTLY from where you left off \u2014 do NOT restart, re-summarize, or repeat any content you already produced. If you were mid-sentence or mid-table, continue that sentence or table. Proceed directly with the next action or output.]",
           metadata: { timestamp: now(), id: randomUUID3() }
         });
       }
@@ -6461,7 +6461,10 @@ ${textContent}` };
         let chunkCount = 0;
         const hasRunTimeout = timeoutMs > 0;
         const streamDeadline = hasRunTimeout ? start + timeoutMs : 0;
+        const hasSoftDeadline = softDeadlineMs > 0;
+        const INTER_CHUNK_TIMEOUT_MS = 6e4;
         const fullStreamIterator = result.fullStream[Symbol.asyncIterator]();
+        let softDeadlineFiredDuringStream = false;
         try {
           while (true) {
             if (isCancelled()) {
@@ -6469,8 +6472,8 @@ ${textContent}` };
               return;
             }
             if (hasRunTimeout) {
-              const remaining2 = streamDeadline - now();
-              if (remaining2 <= 0) {
+              const remaining = streamDeadline - now();
+              if (remaining <= 0) {
                 yield pushEvent({
                   type: "run:error",
                   runId,
@@ -6485,22 +6488,33 @@ ${textContent}` };
                 return;
               }
             }
-            const remaining = hasRunTimeout ? streamDeadline - now() : Infinity;
-            const timeout = chunkCount === 0 ? Math.min(remaining, FIRST_CHUNK_TIMEOUT_MS) : hasRunTimeout ? remaining : 0;
+            if (hasSoftDeadline && chunkCount > 0 && now() - start >= softDeadlineMs) {
+              softDeadlineFiredDuringStream = true;
+              break;
+            }
+            const hardRemaining = hasRunTimeout ? streamDeadline - now() : Infinity;
+            const softRemaining = hasSoftDeadline ? Math.max(0, start + softDeadlineMs - now()) : Infinity;
+            const deadlineRemaining = Math.min(hardRemaining, softRemaining);
+            const timeout = chunkCount === 0 ? Math.min(deadlineRemaining, FIRST_CHUNK_TIMEOUT_MS) : Math.min(deadlineRemaining, INTER_CHUNK_TIMEOUT_MS);
             let nextPart;
-            if (timeout <= 0 && chunkCount > 0) {
+            if (timeout <= 0 && chunkCount > 0 && !hasSoftDeadline) {
               nextPart = await fullStreamIterator.next();
             } else {
+              const effectiveTimeout = Math.max(timeout, 1);
               let timer;
               nextPart = await Promise.race([
                 fullStreamIterator.next(),
                 new Promise((resolve12) => {
-                  timer = setTimeout(() => resolve12(null), timeout);
+                  timer = setTimeout(() => resolve12(null), effectiveTimeout);
                 })
               ]);
               clearTimeout(timer);
             }
             if (nextPart === null) {
+              if (hasSoftDeadline && deadlineRemaining <= INTER_CHUNK_TIMEOUT_MS) {
+                softDeadlineFiredDuringStream = true;
+                break;
+              }
               const isFirstChunk = chunkCount === 0;
               console.error(
                 `[poncho][harness] Stream timeout waiting for ${isFirstChunk ? "first" : "next"} chunk: model="${modelName}", step=${step}, chunks=${chunkCount}, elapsed=${now() - start}ms`
@@ -6533,11 +6547,42 @@ ${textContent}` };
           fullStreamIterator.return?.(void 0)?.catch?.(() => {
           });
         }
+        if (softDeadlineFiredDuringStream) {
+          if (fullText.length > 0) {
+            messages.push({
+              role: "assistant",
+              content: fullText,
+              metadata: { timestamp: now(), id: randomUUID3(), step }
+            });
+          }
+          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
+          };
+          console.info(`[poncho][harness] Soft deadline fired mid-stream at step ${step} (${(now() - start).toFixed(0)}ms). Checkpointing with ${fullText.length} chars of partial text.`);
+          yield pushEvent({ type: "run:completed", runId, result: result_ });
+          return;
+        }
         if (isCancelled()) {
           yield emitCancellation();
           return;
         }
         if (softDeadlineMs > 0 && now() - start > softDeadlineMs) {
+          if (fullText.length > 0) {
+            messages.push({
+              role: "assistant",
+              content: fullText,
+              metadata: { timestamp: now(), id: randomUUID3(), step }
+            });
+          }
           const result_ = {
             status: "completed",
             response: responseText + fullText,
@@ -6789,6 +6834,13 @@ ${textContent}` };
           batchResults = await this.dispatcher.executeBatch(approvedCalls, toolContext);
         }
         if (batchResults === TOOL_DEADLINE_SENTINEL) {
+          if (fullText.length > 0) {
+            messages.push({
+              role: "assistant",
+              content: fullText,
+              metadata: { timestamp: now(), id: randomUUID3(), step }
+            });
+          }
           const result_ = {
             status: "completed",
             response: responseText + fullText,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@poncho-ai/harness",
-  "version": "0.29.0",
+  "version": "0.30.0",
   "description": "Agent execution runtime - conversation loop, tool dispatch, streaming",
   "repository": {
     "type": "git",
@@ -34,7 +34,7 @@
     "redis": "^5.10.0",
     "yaml": "^2.4.0",
     "zod": "^3.22.0",
-    "@poncho-ai/sdk": "1.6.2"
+    "@poncho-ai/sdk": "1.6.3"
   },
   "devDependencies": {
     "@types/mustache": "^4.2.6",

package/src/harness.ts

@@ -1643,7 +1643,7 @@ ${boundedMainMemory.trim()}`
       if (lastMsg && lastMsg.role !== "user") {
         messages.push({
           role: "user",
-          content: "[System: Your previous turn was interrupted by a time limit. Continue from where you left off — do NOT repeat what you already said. Proceed directly with the next action or tool call.]",
+          content: "[System: Your previous turn was interrupted by a time limit. Your partial response above is already visible to the user. Continue EXACTLY from where you left off — do NOT restart, re-summarize, or repeat any content you already produced. If you were mid-sentence or mid-table, continue that sentence or table. Proceed directly with the next action or output.]",
           metadata: { timestamp: now(), id: randomUUID() },
         });
       }
@@ -2048,7 +2048,10 @@ ${boundedMainMemory.trim()}`
         let chunkCount = 0;
         const hasRunTimeout = timeoutMs > 0;
         const streamDeadline = hasRunTimeout ? start + timeoutMs : 0;
+        const hasSoftDeadline = softDeadlineMs > 0;
+        const INTER_CHUNK_TIMEOUT_MS = 60_000;
         const fullStreamIterator = result.fullStream[Symbol.asyncIterator]();
+        let softDeadlineFiredDuringStream = false;
         try {
           while (true) {
             if (isCancelled()) {
@@ -2072,25 +2075,36 @@ ${boundedMainMemory.trim()}`
                 return;
               }
             }
-            const remaining = hasRunTimeout ? streamDeadline - now() : Infinity;
+            if (hasSoftDeadline && chunkCount > 0 && now() - start >= softDeadlineMs) {
+              softDeadlineFiredDuringStream = true;
+              break;
+            }
+            const hardRemaining = hasRunTimeout ? streamDeadline - now() : Infinity;
+            const softRemaining = hasSoftDeadline ? Math.max(0, (start + softDeadlineMs) - now()) : Infinity;
+            const deadlineRemaining = Math.min(hardRemaining, softRemaining);
             const timeout = chunkCount === 0
-              ? Math.min(remaining, FIRST_CHUNK_TIMEOUT_MS)
-              : hasRunTimeout ? remaining : 0;
+              ? Math.min(deadlineRemaining, FIRST_CHUNK_TIMEOUT_MS)
+              : Math.min(deadlineRemaining, INTER_CHUNK_TIMEOUT_MS);
             let nextPart: IteratorResult<(typeof result.fullStream) extends AsyncIterable<infer T> ? T : never> | null;
-            if (timeout <= 0 && chunkCount > 0) {
+            if (timeout <= 0 && chunkCount > 0 && !hasSoftDeadline) {
               nextPart = await fullStreamIterator.next();
             } else {
+              const effectiveTimeout = Math.max(timeout, 1);
               let timer: ReturnType<typeof setTimeout> | undefined;
               nextPart = await Promise.race([
                 fullStreamIterator.next(),
                 new Promise<null>((resolve) => {
-                  timer = setTimeout(() => resolve(null), timeout);
+                  timer = setTimeout(() => resolve(null), effectiveTimeout);
                 }),
               ]);
               clearTimeout(timer);
             }
 
             if (nextPart === null) {
+              if (hasSoftDeadline && deadlineRemaining <= INTER_CHUNK_TIMEOUT_MS) {
+                softDeadlineFiredDuringStream = true;
+                break;
+              }
               const isFirstChunk = chunkCount === 0;
               console.error(
                 `[poncho][harness] Stream timeout waiting for ${isFirstChunk ? "first" : "next"} chunk: model="${modelName}", step=${step}, chunks=${chunkCount}, elapsed=${now() - start}ms`,
@@ -2125,6 +2139,31 @@ ${boundedMainMemory.trim()}`
           fullStreamIterator.return?.(undefined)?.catch?.(() => {});
         }
 
+        if (softDeadlineFiredDuringStream) {
+          if (fullText.length > 0) {
+            messages.push({
+              role: "assistant",
+              content: fullText,
+              metadata: { timestamp: now(), id: randomUUID(), step },
+            });
+          }
+          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,
+          };
+          console.info(`[poncho][harness] Soft deadline fired mid-stream at step ${step} (${(now() - start).toFixed(0)}ms). Checkpointing with ${fullText.length} chars of partial text.`);
+          yield pushEvent({ type: "run:completed", runId, result: result_ });
+          return;
+        }
+
         if (isCancelled()) {
           yield emitCancellation();
           return;
@@ -2133,6 +2172,13 @@ ${boundedMainMemory.trim()}`
       // 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) {
+        if (fullText.length > 0) {
+          messages.push({
+            role: "assistant",
+            content: fullText,
+            metadata: { timestamp: now(), id: randomUUID(), step },
+          });
+        }
         const result_: RunResult = {
           status: "completed",
           response: responseText + fullText,
@@ -2446,6 +2492,13 @@ ${boundedMainMemory.trim()}`
       }
 
       if ((batchResults as unknown) === TOOL_DEADLINE_SENTINEL) {
+        if (fullText.length > 0) {
+          messages.push({
+            role: "assistant",
+            content: fullText,
+            metadata: { timestamp: now(), id: randomUUID(), step },
+          });
+        }
         const result_: RunResult = {
           status: "completed",
           response: responseText + fullText,

package/src/state.ts

@@ -71,6 +71,10 @@ export interface Conversation {
   /** Harness-internal message chain preserved across continuation runs.
    *  Cleared when a run completes without continuation. */
   _continuationMessages?: Message[];
+  /** Number of continuation pickups for the current multi-step run.
+   *  Reset when a run completes without continuation. Used to enforce
+   *  a maximum continuation count across all entry points. */
+  _continuationCount?: number;
   /** Full structured message chain from the last harness run, including
    *  tool-call and tool-result messages the model needs for context.
    *  Unlike `_continuationMessages`, this is always set after a run

package/.turbo/turbo-lint.log

@@ -1,6 +0,0 @@
-
-> @poncho-ai/harness@0.11.2 lint /Users/cesar/Dev/latitude/poncho-ai/packages/harness
-> eslint src/
-
-sh: eslint: command not found
- ELIFECYCLE  Command failed.

package/.turbo/turbo-test.log

@@ -1,34 +0,0 @@
-
-> @poncho-ai/harness@0.26.0 test /Users/cesar/Dev/latitude/poncho-ai/packages/harness
-> vitest
-
-
- RUN  v1.6.1 /Users/cesar/Dev/latitude/poncho-ai/packages/harness
-
-stdout | test/mcp.test.ts > mcp bridge protocol transports > discovers and calls tools over streamable HTTP
-[poncho][mcp] {"event":"catalog.loaded","server":"remote","discoveredCount":1}
-[poncho][mcp] {"event":"tools.selected","requestedPatternCount":1,"registeredCount":1,"filteredByPolicyCount":0,"filteredByIntentCount":0}
-
-stdout | test/mcp.test.ts > mcp bridge protocol transports > sends custom headers alongside bearer token
-[poncho][mcp] {"event":"catalog.loaded","server":"custom-headers","discoveredCount":1}
-
-stderr | test/mcp.test.ts > mcp bridge protocol transports > skips discovery when bearer token env value is missing
-stdout | test/mcp.test.ts > mcp bridge protocol transports > selects discovered tools by requested patterns
-[poncho][mcp] {"event":"auth.token_missing","server":"remote","tokenEnv":"MISSING_TOKEN_ENV"}
-[poncho][mcp] {"event":"catalog.loaded","server":"remote","discoveredCount":2}
-[poncho][mcp] {"event":"tools.selected","requestedPatternCount":1,"registeredCount":1,"filteredByPolicyCount":0,"filteredByIntentCount":1}
-
-[poncho][mcp] {"event":"tools.selected","requestedPatternCount":1,"registeredCount":2,"filteredByPolicyCount":0,"filteredByIntentCount":0}
-
-stdout | test/mcp.test.ts > mcp bridge protocol transports > skips discovery when bearer token env value is missing
-[poncho][mcp] {"event":"tools.selected","requestedPatternCount":1,"registeredCount":0,"filteredByPolicyCount":0,"filteredByIntentCount":0}
-
-[event] step:completed {"type":"step:completed","step":1,"duration":1}
- ✓ test/telemetry.test.ts  (3 tests) 5ms
-[event] step:started {"type":"step:started","step":2}
- ✓ test/schema-converter.test.ts  (27 tests) 13ms
-stdout | test/mcp.test.ts > mcp bridge protocol transports > returns actionable errors for 403 permission failures
-[poncho][mcp] {"event":"catalog.loaded","server":"remote","discoveredCount":1}
-[poncho][mcp] {"event":"tools.selected","requestedPatternCount":1,"registeredCount":1,"filteredByPolicyCount":0,"filteredByIntentCount":0}
-
- ✓ test/mcp.test.ts  (7 tests) 84ms