@poncho-ai/harness 0.17.0 → 0.18.0

package/.turbo/turbo-build.log

@@ -1,5 +1,5 @@
 
-> @poncho-ai/harness@0.17.0 build /home/runner/work/poncho-ai/poncho-ai/packages/harness
+> @poncho-ai/harness@0.18.0 build /home/runner/work/poncho-ai/poncho-ai/packages/harness
 > tsup src/index.ts --format esm --dts
 
 CLI Building entry: src/index.ts
@@ -7,8 +7,8 @@
 CLI tsup v8.5.1
 CLI Target: es2022
 ESM Build start
-ESM dist/index.js 196.31 KB
-ESM ⚡️ Build success in 155ms
+ESM dist/index.js 196.78 KB
+ESM ⚡️ Build success in 133ms
 DTS Build start
-DTS ⚡️ Build success in 6569ms
-DTS dist/index.d.ts 23.97 KB
+DTS ⚡️ Build success in 6813ms
+DTS dist/index.d.ts 24.01 KB

package/CHANGELOG.md

@@ -1,5 +1,19 @@
 # @poncho-ai/harness
 
+## 0.18.0
+
+### Minor Changes
+
+- [`cd6ccd7`](https://github.com/cesr/poncho-ai/commit/cd6ccd7846e16fbaf17167617666796320ec29ce) Thanks [@cesr](https://github.com/cesr)! - Add MCP custom headers support, tool:generating streaming feedback, and cross-owner subagent recovery
+  - **MCP custom headers**: `poncho mcp add --header "Name: value"` and `headers` config field let servers like Arcade receive extra HTTP headers alongside bearer auth.
+  - **tool:generating event**: the harness now emits `tool:generating` events when the model begins writing tool-call arguments, so the web UI shows real-time "preparing <tool>" feedback instead of appearing stuck during large tool calls.
+  - **Subagent recovery**: `list`/`listSummaries` accept optional `ownerId` so stale-subagent recovery on server restart scans across all owners.
+
+### Patch Changes
+
+- Updated dependencies [[`cd6ccd7`](https://github.com/cesr/poncho-ai/commit/cd6ccd7846e16fbaf17167617666796320ec29ce)]:
+  - @poncho-ai/sdk@1.3.0
+
 ## 0.17.0
 
 ### Minor Changes

package/dist/index.d.ts

@@ -205,6 +205,7 @@ interface RemoteMcpServerConfig {
         type: "bearer";
         tokenEnv?: string;
     };
+    headers?: Record<string, string>;
     timeoutMs?: number;
     reconnectAttempts?: number;
     reconnectDelayMs?: number;

package/dist/index.js

@@ -1306,16 +1306,19 @@ var StreamableHttpMcpRpcClient = class {
   endpoint;
   timeoutMs;
   bearerToken;
+  customHeaders;
   idCounter = 1;
   initialized = false;
   sessionId;
-  constructor(endpoint, timeoutMs = 1e4, bearerToken) {
+  constructor(endpoint, timeoutMs = 1e4, bearerToken, customHeaders) {
     this.endpoint = endpoint;
     this.timeoutMs = timeoutMs;
     this.bearerToken = bearerToken;
+    this.customHeaders = customHeaders ?? {};
   }
   buildHeaders(accept) {
     const headers = {
+      ...this.customHeaders,
       "Content-Type": "application/json",
       Accept: accept
     };
@@ -1596,7 +1599,8 @@ var LocalMcpBridge = class {
         new StreamableHttpMcpRpcClient(
           server.url,
           server.timeoutMs ?? 1e4,
-          server.auth?.tokenEnv ? process.env[server.auth.tokenEnv] : void 0
+          server.auth?.tokenEnv ? process.env[server.auth.tokenEnv] : void 0,
+          server.headers
         )
       );
     }
@@ -4063,7 +4067,7 @@ ${textContent}` };
         let chunkCount = 0;
         const hasRunTimeout = timeoutMs > 0;
         const streamDeadline = hasRunTimeout ? start + timeoutMs : 0;
-        const textIterator = result.textStream[Symbol.asyncIterator]();
+        const fullStreamIterator = result.fullStream[Symbol.asyncIterator]();
         try {
           while (true) {
             if (isCancelled()) {
@@ -4089,20 +4093,20 @@ ${textContent}` };
             }
             const remaining = hasRunTimeout ? streamDeadline - now() : Infinity;
             const timeout = chunkCount === 0 ? Math.min(remaining, FIRST_CHUNK_TIMEOUT_MS) : hasRunTimeout ? remaining : 0;
-            let nextChunk;
+            let nextPart;
             if (timeout <= 0 && chunkCount > 0) {
-              nextChunk = await textIterator.next();
+              nextPart = await fullStreamIterator.next();
             } else {
               let timer;
-              nextChunk = await Promise.race([
-                textIterator.next(),
+              nextPart = await Promise.race([
+                fullStreamIterator.next(),
                 new Promise((resolve10) => {
                   timer = setTimeout(() => resolve10(null), timeout);
                 })
               ]);
               clearTimeout(timer);
             }
-            if (nextChunk === null) {
+            if (nextPart === null) {
               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`
@@ -4120,13 +4124,19 @@ ${textContent}` };
               });
               return;
             }
-            if (nextChunk.done) break;
-            chunkCount += 1;
-            fullText += nextChunk.value;
-            yield pushEvent({ type: "model:chunk", content: nextChunk.value });
+            if (nextPart.done) break;
+            const part = nextPart.value;
+            if (part.type === "text-delta") {
+              chunkCount += 1;
+              fullText += part.text;
+              yield pushEvent({ type: "model:chunk", content: part.text });
+            } else if (part.type === "tool-input-start") {
+              chunkCount += 1;
+              yield pushEvent({ type: "tool:generating", tool: part.toolName, toolCallId: part.id });
+            }
           }
         } finally {
-          textIterator.return?.(void 0)?.catch?.(() => {
+          fullStreamIterator.return?.(void 0)?.catch?.(() => {
           });
         }
         if (isCancelled()) {
@@ -4693,13 +4703,13 @@ var InMemoryConversationStore = class {
       }
     }
   }
-  async list(ownerId = DEFAULT_OWNER) {
+  async list(ownerId) {
     this.purgeExpired();
-    return Array.from(this.conversations.values()).filter((conversation) => conversation.ownerId === ownerId).sort((a, b) => b.updatedAt - a.updatedAt);
+    return Array.from(this.conversations.values()).filter((conversation) => !ownerId || conversation.ownerId === ownerId).sort((a, b) => b.updatedAt - a.updatedAt);
   }
-  async listSummaries(ownerId = DEFAULT_OWNER) {
+  async listSummaries(ownerId) {
     this.purgeExpired();
-    return Array.from(this.conversations.values()).filter((c) => c.ownerId === ownerId).sort((a, b) => b.updatedAt - a.updatedAt).map((c) => ({
+    return Array.from(this.conversations.values()).filter((c) => !ownerId || c.ownerId === ownerId).sort((a, b) => b.updatedAt - a.updatedAt).map((c) => ({
       conversationId: c.conversationId,
       title: c.title,
       updatedAt: c.updatedAt,
@@ -4875,9 +4885,9 @@ var FileConversationStore = class {
     });
     await this.writing;
   }
-  async list(ownerId = DEFAULT_OWNER) {
+  async list(ownerId) {
     await this.ensureLoaded();
-    const summaries = Array.from(this.conversations.values()).filter((conversation) => conversation.ownerId === ownerId).sort((a, b) => b.updatedAt - a.updatedAt);
+    const summaries = Array.from(this.conversations.values()).filter((conversation) => !ownerId || conversation.ownerId === ownerId).sort((a, b) => b.updatedAt - a.updatedAt);
     const conversations = [];
     for (const summary of summaries) {
       const loaded = await this.readConversationFile(summary.fileName);
@@ -4887,9 +4897,9 @@ var FileConversationStore = class {
     }
     return conversations;
   }
-  async listSummaries(ownerId = DEFAULT_OWNER) {
+  async listSummaries(ownerId) {
     await this.ensureLoaded();
-    return Array.from(this.conversations.values()).filter((c) => c.ownerId === ownerId).sort((a, b) => b.updatedAt - a.updatedAt).map((c) => ({
+    return Array.from(this.conversations.values()).filter((c) => !ownerId || c.ownerId === ownerId).sort((a, b) => b.updatedAt - a.updatedAt).map((c) => ({
       conversationId: c.conversationId,
       title: c.title,
       updatedAt: c.updatedAt,
@@ -5115,11 +5125,14 @@ var KeyValueConversationStoreBase = class {
       return void 0;
     }
   }
-  async list(ownerId = DEFAULT_OWNER) {
+  async list(ownerId) {
     const kv = await this.client();
     if (!kv) {
       return await this.memoryFallback.list(ownerId);
     }
+    if (!ownerId) {
+      return [];
+    }
     const ids = await this.getOwnerConversationIds(ownerId);
     const conversations = [];
     for (const id of ids) {
@@ -5134,11 +5147,14 @@ var KeyValueConversationStoreBase = class {
     }
     return conversations.sort((a, b) => b.updatedAt - a.updatedAt);
   }
-  async listSummaries(ownerId = DEFAULT_OWNER) {
+  async listSummaries(ownerId) {
     const kv = await this.client();
     if (!kv) {
       return await this.memoryFallback.listSummaries(ownerId);
     }
+    if (!ownerId) {
+      return [];
+    }
     const ids = await this.getOwnerConversationIds(ownerId);
     const summaries = [];
     for (const id of ids) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@poncho-ai/harness",
-  "version": "0.17.0",
+  "version": "0.18.0",
   "description": "Agent execution runtime - conversation loop, tool dispatch, streaming",
   "repository": {
     "type": "git",
@@ -31,7 +31,7 @@
     "redis": "^5.10.0",
     "yaml": "^2.4.0",
     "zod": "^3.22.0",
-    "@poncho-ai/sdk": "1.2.0"
+    "@poncho-ai/sdk": "1.3.0"
   },
   "devDependencies": {
     "@types/mustache": "^4.2.6",

package/src/harness.ts

@@ -1682,15 +1682,14 @@ ${boundedMainMemory.trim()}`
             isEnabled: telemetryEnabled && !!this.latitudeTelemetry,
           },
         });
-        // Stream text chunks — enforce overall run timeout per chunk.
-        // The top-of-step timeout check cannot fire while we are
-        // blocked inside the textStream async iterator, so we race
-        // each next() call against the remaining time budget.
+        // Stream full response — use fullStream to get visibility into
+        // tool-call generation (tool-input-start) in addition to text deltas.
+        // Enforce overall run timeout per part.
         let fullText = "";
         let chunkCount = 0;
         const hasRunTimeout = timeoutMs > 0;
         const streamDeadline = hasRunTimeout ? start + timeoutMs : 0;
-        const textIterator = result.textStream[Symbol.asyncIterator]();
+        const fullStreamIterator = result.fullStream[Symbol.asyncIterator]();
         try {
           while (true) {
             if (isCancelled()) {
@@ -1714,21 +1713,17 @@ ${boundedMainMemory.trim()}`
                 return;
               }
             }
-            // Use a shorter timeout for the first chunk to detect
-            // non-responsive models quickly instead of waiting minutes.
-            // When no run timeout is set, only the first chunk is time-bounded.
             const remaining = hasRunTimeout ? streamDeadline - now() : Infinity;
             const timeout = chunkCount === 0
               ? Math.min(remaining, FIRST_CHUNK_TIMEOUT_MS)
               : hasRunTimeout ? remaining : 0;
-            let nextChunk: IteratorResult<string> | null;
+            let nextPart: IteratorResult<(typeof result.fullStream) extends AsyncIterable<infer T> ? T : never> | null;
             if (timeout <= 0 && chunkCount > 0) {
-              // No time budget — await the stream directly (development mode, no run timeout)
-              nextChunk = await textIterator.next();
+              nextPart = await fullStreamIterator.next();
             } else {
               let timer: ReturnType<typeof setTimeout> | undefined;
-              nextChunk = await Promise.race([
-                textIterator.next(),
+              nextPart = await Promise.race([
+                fullStreamIterator.next(),
                 new Promise<null>((resolve) => {
                   timer = setTimeout(() => resolve(null), timeout);
                 }),
@@ -1736,16 +1731,14 @@ ${boundedMainMemory.trim()}`
               clearTimeout(timer);
             }
 
-            if (nextChunk === null) {
+            if (nextPart === null) {
               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`,
               );
               if (isFirstChunk) {
-                // Throw so the step-level retry logic can handle it.
                 throw new FirstChunkTimeoutError(modelName, FIRST_CHUNK_TIMEOUT_MS);
               }
-              // Mid-stream timeout: not retryable (partial response would be lost)
               yield pushEvent({
                 type: "run:error",
                 runId,
@@ -1757,14 +1750,20 @@ ${boundedMainMemory.trim()}`
               return;
             }
 
-            if (nextChunk.done) break;
-            chunkCount += 1;
-            fullText += nextChunk.value;
-            yield pushEvent({ type: "model:chunk", content: nextChunk.value });
+            if (nextPart.done) break;
+            const part = nextPart.value;
+
+            if (part.type === "text-delta") {
+              chunkCount += 1;
+              fullText += part.text;
+              yield pushEvent({ type: "model:chunk", content: part.text });
+            } else if (part.type === "tool-input-start") {
+              chunkCount += 1;
+              yield pushEvent({ type: "tool:generating", tool: part.toolName, toolCallId: part.id });
+            }
           }
         } finally {
-          // Best-effort cleanup of the underlying stream/connection.
-          textIterator.return?.(undefined)?.catch?.(() => {});
+          fullStreamIterator.return?.(undefined)?.catch?.(() => {});
         }
 
         if (isCancelled()) {
@@ -1773,8 +1772,6 @@ ${boundedMainMemory.trim()}`
         }
 
       // Check finish reason for error / abnormal completions.
-      // textStream silently swallows model-level errors – they only
-      // surface through finishReason (or fullStream, which we don't use).
       const finishReason = await result.finishReason;
 
       if (finishReason === "error") {

package/src/mcp.ts

@@ -12,6 +12,7 @@ export interface RemoteMcpServerConfig {
     type: "bearer";
     tokenEnv?: string;
   };
+  headers?: Record<string, string>;
   timeoutMs?: number;
   reconnectAttempts?: number;
   reconnectDelayMs?: number;
@@ -46,18 +47,26 @@ class StreamableHttpMcpRpcClient implements McpRpcClient {
   private readonly endpoint: string;
   private readonly timeoutMs: number;
   private readonly bearerToken?: string;
+  private readonly customHeaders: Record<string, string>;
   private idCounter = 1;
   private initialized = false;
   private sessionId?: string;
 
-  constructor(endpoint: string, timeoutMs = 10_000, bearerToken?: string) {
+  constructor(
+    endpoint: string,
+    timeoutMs = 10_000,
+    bearerToken?: string,
+    customHeaders?: Record<string, string>,
+  ) {
     this.endpoint = endpoint;
     this.timeoutMs = timeoutMs;
     this.bearerToken = bearerToken;
+    this.customHeaders = customHeaders ?? {};
   }
 
   private buildHeaders(accept: string): Record<string, string> {
     const headers: Record<string, string> = {
+      ...this.customHeaders,
       "Content-Type": "application/json",
       Accept: accept,
     };
@@ -367,6 +376,7 @@ export class LocalMcpBridge {
           server.url,
           server.timeoutMs ?? 10_000,
           server.auth?.tokenEnv ? process.env[server.auth.tokenEnv] : undefined,
+          server.headers,
         ),
       );
     }

package/src/state.ts

@@ -223,17 +223,17 @@ export class InMemoryConversationStore implements ConversationStore {
     }
   }
 
-  async list(ownerId = DEFAULT_OWNER): Promise<Conversation[]> {
+  async list(ownerId?: string): Promise<Conversation[]> {
     this.purgeExpired();
     return Array.from(this.conversations.values())
-      .filter((conversation) => conversation.ownerId === ownerId)
+      .filter((conversation) => !ownerId || conversation.ownerId === ownerId)
       .sort((a, b) => b.updatedAt - a.updatedAt);
   }
 
-  async listSummaries(ownerId = DEFAULT_OWNER): Promise<ConversationSummary[]> {
+  async listSummaries(ownerId?: string): Promise<ConversationSummary[]> {
     this.purgeExpired();
     return Array.from(this.conversations.values())
-      .filter((c) => c.ownerId === ownerId)
+      .filter((c) => !ownerId || c.ownerId === ownerId)
       .sort((a, b) => b.updatedAt - a.updatedAt)
       .map((c) => ({
         conversationId: c.conversationId,
@@ -454,10 +454,10 @@ class FileConversationStore implements ConversationStore {
     await this.writing;
   }
 
-  async list(ownerId = DEFAULT_OWNER): Promise<Conversation[]> {
+  async list(ownerId?: string): Promise<Conversation[]> {
     await this.ensureLoaded();
     const summaries = Array.from(this.conversations.values())
-      .filter((conversation) => conversation.ownerId === ownerId)
+      .filter((conversation) => !ownerId || conversation.ownerId === ownerId)
       .sort((a, b) => b.updatedAt - a.updatedAt);
     const conversations: Conversation[] = [];
     for (const summary of summaries) {
@@ -469,10 +469,10 @@ class FileConversationStore implements ConversationStore {
     return conversations;
   }
 
-  async listSummaries(ownerId = DEFAULT_OWNER): Promise<ConversationSummary[]> {
+  async listSummaries(ownerId?: string): Promise<ConversationSummary[]> {
     await this.ensureLoaded();
     return Array.from(this.conversations.values())
-      .filter((c) => c.ownerId === ownerId)
+      .filter((c) => !ownerId || c.ownerId === ownerId)
       .sort((a, b) => b.updatedAt - a.updatedAt)
       .map((c) => ({
         conversationId: c.conversationId,
@@ -751,11 +751,15 @@ abstract class KeyValueConversationStoreBase implements ConversationStore {
     }
   }
 
-  async list(ownerId = DEFAULT_OWNER): Promise<Conversation[]> {
+  async list(ownerId?: string): Promise<Conversation[]> {
     const kv = await this.client();
     if (!kv) {
       return await this.memoryFallback.list(ownerId);
     }
+    if (!ownerId) {
+      // KV stores index per-owner; cross-owner listing not supported
+      return [];
+    }
     const ids = await this.getOwnerConversationIds(ownerId);
     const conversations: Conversation[] = [];
     for (const id of ids) {
@@ -772,11 +776,14 @@ abstract class KeyValueConversationStoreBase implements ConversationStore {
     return conversations.sort((a, b) => b.updatedAt - a.updatedAt);
   }
 
-  async listSummaries(ownerId = DEFAULT_OWNER): Promise<ConversationSummary[]> {
+  async listSummaries(ownerId?: string): Promise<ConversationSummary[]> {
     const kv = await this.client();
     if (!kv) {
       return await this.memoryFallback.listSummaries(ownerId);
     }
+    if (!ownerId) {
+      return [];
+    }
     const ids = await this.getOwnerConversationIds(ownerId);
     const summaries: ConversationSummary[] = [];
     for (const id of ids) {

package/test/mcp.test.ts

@@ -126,6 +126,89 @@ describe("mcp bridge protocol transports", () => {
     expect(deleteSeen).toBe(true);
   });
 
+  it("sends custom headers alongside bearer token", async () => {
+    process.env.LINEAR_TOKEN = "token-123";
+    let capturedHeaders: Record<string, string | undefined> = {};
+    const server = createServer(async (req, res) => {
+      if (req.method === "DELETE") {
+        res.statusCode = 200;
+        res.end();
+        return;
+      }
+      const chunks: Buffer[] = [];
+      for await (const chunk of req) chunks.push(Buffer.from(chunk));
+      const body = Buffer.concat(chunks).toString("utf8");
+      const payload = body.trim().length > 0 ? (JSON.parse(body) as any) : {};
+      capturedHeaders = {
+        authorization: req.headers.authorization,
+        "x-custom-id": req.headers["x-custom-id"] as string | undefined,
+        "x-another": req.headers["x-another"] as string | undefined,
+      };
+      if (payload.method === "initialize") {
+        res.setHeader("Content-Type", "application/json");
+        res.setHeader("Mcp-Session-Id", "s");
+        res.end(
+          JSON.stringify({
+            jsonrpc: "2.0",
+            id: payload.id,
+            result: {
+              protocolVersion: "2025-03-26",
+              capabilities: { tools: { listChanged: true } },
+              serverInfo: { name: "remote", version: "1.0.0" },
+            },
+          }),
+        );
+        return;
+      }
+      if (payload.method === "notifications/initialized") {
+        res.statusCode = 202;
+        res.end();
+        return;
+      }
+      if (payload.method === "tools/list") {
+        res.setHeader("Content-Type", "application/json");
+        res.end(
+          JSON.stringify({
+            jsonrpc: "2.0",
+            id: payload.id,
+            result: {
+              tools: [
+                { name: "ping", inputSchema: { type: "object", properties: {} } },
+              ],
+            },
+          }),
+        );
+        return;
+      }
+      res.statusCode = 404;
+      res.end();
+    });
+    await new Promise<void>((r) => server.listen(0, () => r()));
+    const address = server.address();
+    if (!address || typeof address === "string") throw new Error("Unexpected address");
+
+    const bridge = new LocalMcpBridge({
+      mcp: [
+        {
+          name: "custom-headers",
+          url: `http://127.0.0.1:${address.port}/mcp`,
+          auth: { type: "bearer", tokenEnv: "LINEAR_TOKEN" },
+          headers: { "X-Custom-Id": "user-42", "X-Another": "value" },
+        },
+      ],
+    });
+
+    await bridge.startLocalServers();
+    await bridge.discoverTools();
+
+    expect(capturedHeaders.authorization).toBe("Bearer token-123");
+    expect(capturedHeaders["x-custom-id"]).toBe("user-42");
+    expect(capturedHeaders["x-another"]).toBe("value");
+
+    await bridge.stopLocalServers();
+    await new Promise<void>((r) => server.close(() => r()));
+  });
+
   it("fails fast on duplicate server names", () => {
     expect(
       () =>