@poncho-ai/sdk 1.5.0 → 1.6.1

package/.turbo/turbo-build.log

@@ -1,5 +1,5 @@
 
-> @poncho-ai/sdk@1.5.0 build /home/runner/work/poncho-ai/poncho-ai/packages/sdk
+> @poncho-ai/sdk@1.6.1 build /home/runner/work/poncho-ai/poncho-ai/packages/sdk
 > tsup src/index.ts --format esm --dts
 
 CLI Building entry: src/index.ts
@@ -8,7 +8,7 @@
 CLI Target: es2022
 ESM Build start
 ESM dist/index.js 11.44 KB
-ESM ⚡️ Build success in 20ms
+ESM ⚡️ Build success in 17ms
 DTS Build start
-DTS ⚡️ Build success in 1276ms
-DTS dist/index.d.ts 21.84 KB
+DTS ⚡️ Build success in 1248ms
+DTS dist/index.d.ts 22.39 KB

package/CHANGELOG.md

@@ -1,5 +1,30 @@
 # @poncho-ai/sdk
 
+## 1.6.1
+
+### 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.
+
+## 1.6.0
+
+### Minor Changes
+
+- [#42](https://github.com/cesr/poncho-ai/pull/42) [`e58a984`](https://github.com/cesr/poncho-ai/commit/e58a984efaa673b649318102bbf735fb4c2f9172) Thanks [@cesr](https://github.com/cesr)! - Add continuation model and fire-and-forget subagents
+
+  **Continuation model**: Agents no longer send a synthetic `"Continue"` user message between steps. Instead, the harness injects a transient signal when needed, and the full internal message chain is preserved across continuations so the LLM never loses context. `RunInput` gains `disableSoftDeadline` and `RunResult` gains `continuationMessages`.
+
+  **Fire-and-forget subagents**: Subagents now run asynchronously in the background. `spawn_subagent` returns immediately with a subagent ID; results are delivered back to the parent conversation as a callback once the subagent completes. Subagents cannot spawn their own subagents. The web UI shows results in a collapsible disclosure and reconnects the live event stream automatically when the parent agent resumes.
+
+  **Bug fixes**:
+  - Fixed a race condition where concurrent runs on the same harness instance could assign a subagent or browser tab to the wrong parent conversation (shared `_currentRunConversationId` field replaced with per-run `ToolContext.conversationId`).
+  - Fixed Upstash KV store silently dropping large values by switching from URL-path encoding to request body format for `SET`/`SETEX` commands.
+  - Fixed empty assistant content blocks causing Anthropic `text content blocks must be non-empty` errors.
+
+  **Client**: Added `getConversationStatus()` and `waitForSubagents` option on `sendMessage()`.
+
 ## 1.5.0
 
 ### Minor Changes

package/dist/index.d.ts

@@ -570,6 +570,8 @@ interface RunInput {
     abortSignal?: AbortSignal;
     /** When set, Latitude telemetry groups all turns in this conversation under a single trace. */
     conversationId?: string;
+    /** When true, ignores PONCHO_MAX_DURATION soft deadline (used for background subagent runs). */
+    disableSoftDeadline?: boolean;
 }
 interface TokenUsage {
     input: number;
@@ -583,7 +585,13 @@ interface RunResult {
     tokens: TokenUsage;
     duration: number;
     continuation?: boolean;
+    /** Full message chain from the harness run (populated when continuation=true). */
+    continuationMessages?: Message[];
     maxSteps?: number;
+    /** Estimated current context usage in tokens at end of run (last model input + tool output estimates, reset on compaction). */
+    contextTokens?: number;
+    /** Model context window size in tokens. */
+    contextWindow?: number;
 }
 interface AgentFailure {
     code: string;
@@ -599,6 +607,7 @@ type AgentEvent = {
     type: "run:completed";
     runId: string;
     result: RunResult;
+    pendingSubagents?: boolean;
 } | {
     type: "run:cancelled";
     runId: string;
@@ -712,6 +721,8 @@ type AgentEvent = {
     messagesBefore: number;
     messagesAfter: number;
     compactedMessages?: Message[];
+} | {
+    type: "subagents:pending";
 } | {
     type: "compaction:warning";
     reason: string;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@poncho-ai/sdk",
-  "version": "1.5.0",
+  "version": "1.6.1",
   "description": "Core types and utilities for building Poncho skills",
   "repository": {
     "type": "git",

package/src/index.ts

@@ -102,6 +102,8 @@ export interface RunInput {
   abortSignal?: AbortSignal;
   /** When set, Latitude telemetry groups all turns in this conversation under a single trace. */
   conversationId?: string;
+  /** When true, ignores PONCHO_MAX_DURATION soft deadline (used for background subagent runs). */
+  disableSoftDeadline?: boolean;
 }
 
 export interface TokenUsage {
@@ -117,7 +119,13 @@ export interface RunResult {
   tokens: TokenUsage;
   duration: number;
   continuation?: boolean;
+  /** Full message chain from the harness run (populated when continuation=true). */
+  continuationMessages?: Message[];
   maxSteps?: number;
+  /** Estimated current context usage in tokens at end of run (last model input + tool output estimates, reset on compaction). */
+  contextTokens?: number;
+  /** Model context window size in tokens. */
+  contextWindow?: number;
 }
 
 export interface AgentFailure {
@@ -128,7 +136,7 @@ export interface AgentFailure {
 
 export type AgentEvent =
   | { type: "run:started"; runId: string; agentId: string; contextWindow?: number }
-  | { type: "run:completed"; runId: string; result: RunResult }
+  | { type: "run:completed"; runId: string; result: RunResult; pendingSubagents?: boolean }
   | { type: "run:cancelled"; runId: string }
   | { type: "run:error"; runId: string; error: AgentFailure }
   | { type: "step:started"; step: number }
@@ -187,4 +195,5 @@ export type AgentEvent =
       messagesAfter: number;
       compactedMessages?: Message[];
     }
+  | { type: "subagents:pending" }
   | { type: "compaction:warning"; reason: string };