@sweny-ai/core 0.1.36 → 0.1.38

package/dist/__tests__/claude.test.js

@@ -250,6 +250,105 @@ describe("ClaudeClient", () => {
         expect(result.status).toBe("failed");
         expect(result.data.error).toContain("Too many turns");
     });
+    // terminal_reason was added in @anthropic-ai/claude-agent-sdk v0.2.91.
+    // When the turn budget is exhausted the SDK now emits subtype='success'
+    // alongside terminal_reason='max_turns' instead of the old error subtype,
+    // so the claude.ts handler has to look at both fields or silently accept
+    // truncated output as a success.
+    describe("terminal_reason handling (claude-agent-sdk v0.2.91+)", () => {
+        it("returns failed when terminal_reason is max_turns", async () => {
+            mockQuery.mockReturnValueOnce(makeStream([
+                {
+                    type: "result",
+                    subtype: "success",
+                    result: "partial output",
+                    terminal_reason: "max_turns",
+                },
+            ]));
+            const client = new ClaudeClient();
+            const result = await client.run({ instruction: "x", context: {}, tools: [] });
+            expect(result.status).toBe("failed");
+            expect(result.data.error).toContain("max_turns");
+            expect(result.data.error).toContain("terminated early");
+        });
+        it("returns failed for any non-completed terminal_reason", async () => {
+            // Future SDK versions may add new terminal reasons (aborted_tools,
+            // blocking_limit, etc.) — treat anything that isn't 'completed' as a
+            // failure rather than assume it's recoverable.
+            mockQuery.mockReturnValueOnce(makeStream([
+                {
+                    type: "result",
+                    subtype: "success",
+                    result: "partial",
+                    terminal_reason: "aborted_tools",
+                },
+            ]));
+            const client = new ClaudeClient();
+            const result = await client.run({ instruction: "x", context: {}, tools: [] });
+            expect(result.status).toBe("failed");
+            expect(result.data.error).toContain("aborted_tools");
+        });
+        it("returns success when terminal_reason is 'completed'", async () => {
+            mockQuery.mockReturnValueOnce(makeStream([
+                {
+                    type: "result",
+                    subtype: "success",
+                    result: '{"ok": true}',
+                    terminal_reason: "completed",
+                },
+            ]));
+            const client = new ClaudeClient();
+            const result = await client.run({ instruction: "x", context: {}, tools: [] });
+            expect(result.status).toBe("success");
+            expect(result.data.ok).toBe(true);
+        });
+        it("returns success when terminal_reason is absent (pre-v0.2.91 SDK)", async () => {
+            // Old SDKs never set terminal_reason. The guard must not treat an
+            // absent field as a failure or every caller would break on downgrade.
+            mockQuery.mockReturnValueOnce(makeStream([
+                {
+                    type: "result",
+                    subtype: "success",
+                    result: '{"done": true}',
+                    // terminal_reason deliberately omitted
+                },
+            ]));
+            const client = new ClaudeClient();
+            const result = await client.run({ instruction: "x", context: {}, tools: [] });
+            expect(result.status).toBe("success");
+            expect(result.data.done).toBe(true);
+        });
+        it("preserves toolCalls when failing on terminal_reason", async () => {
+            // A max_turns failure should still surface any tool calls that
+            // happened before the budget ran out, so debuggers can see what
+            // Claude was trying to do.
+            mockQuery.mockReturnValueOnce(makeStream([
+                {
+                    type: "assistant",
+                    message: {
+                        content: [
+                            {
+                                type: "tool_use",
+                                name: "search_code",
+                                input: { query: "foo" },
+                            },
+                        ],
+                    },
+                },
+                {
+                    type: "result",
+                    subtype: "success",
+                    result: "partial",
+                    terminal_reason: "max_turns",
+                },
+            ]));
+            const client = new ClaudeClient();
+            const result = await client.run({ instruction: "x", context: {}, tools: [] });
+            expect(result.status).toBe("failed");
+            expect(result.toolCalls).toHaveLength(1);
+            expect(result.toolCalls[0].tool).toBe("search_code");
+        });
+    });
     it("passes model option to query", async () => {
         mockQuery.mockReturnValueOnce(makeStream([{ type: "result", subtype: "success", result: "ok" }]));
         const client = new ClaudeClient({ model: "claude-sonnet-4-20250514" });

package/dist/claude.js

@@ -112,6 +112,19 @@ export class ClaudeClient {
                 else if (message.type === "result") {
                     const resultMsg = message;
                     if (resultMsg.subtype === "success" && "result" in resultMsg) {
+                        // terminal_reason was added in @anthropic-ai/claude-agent-sdk v0.2.91.
+                        // When the turn budget is exhausted the SDK still emits subtype='success'
+                        // but sets terminal_reason='max_turns'. Treat anything other than
+                        // 'completed' (or absent) as a failure so callers are not silently fed
+                        // incomplete output.
+                        const terminalReason = resultMsg.terminal_reason;
+                        if (terminalReason && terminalReason !== "completed") {
+                            return {
+                                status: "failed",
+                                data: { error: `Claude query terminated early: ${terminalReason}` },
+                                toolCalls,
+                            };
+                        }
                         response = resultMsg.result;
                     }
                     else if ("errors" in resultMsg) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@sweny-ai/core",
-  "version": "0.1.36",
+  "version": "0.1.38",
   "type": "module",
   "bin": {
     "sweny": "./dist/cli/main.js"
@@ -48,7 +48,7 @@
     "test:integration": "vitest run --config vitest.integration.config.ts"
   },
   "dependencies": {
-    "@anthropic-ai/claude-agent-sdk": "^0.2.84",
+    "@anthropic-ai/claude-agent-sdk": "^0.2.92",
     "@clack/prompts": "^1.2.0",
     "chalk": "^5",
     "commander": "^13",