crewly 1.20.349 → 1.20.351

package/frontend/dist/index.html

@@ -10,7 +10,7 @@
     <script>(function(){try{var m=null;try{m=window.localStorage.getItem('crewly-theme')}catch(e){}var d=false;try{d=!!(window.matchMedia&&window.matchMedia('(prefers-color-scheme: dark)').matches)}catch(e){}var t=(m==='light'||m==='dark')?m:(d?'dark':'light');document.documentElement.setAttribute('data-theme',t)}catch(e){}})();</script>
     <!-- Nunito font is self-hosted via @fontsource/nunito (imported in main.tsx) -->
     <title>Crewly AI Studio</title>
-    <script type="module" crossorigin src="/assets/index-36b1788c.js"></script>
+    <script type="module" crossorigin src="/assets/index-b1fc60ea.js"></script>
     <link rel="stylesheet" href="/assets/index-b62e5501.css">
   </head>
   <body class="bg-background-dark font-display text-text-primary-dark">

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "crewly",
-  "version": "1.20.349",
+  "version": "1.20.351",
   "type": "module",
   "description": "Multi-agent orchestration platform for AI coding teams — coordinates Claude Code, Gemini CLI, and Codex agents with a real-time web dashboard",
   "workspaces": [

package/packages/crewly-agent/src/runtime/tool-registry.test.ts

@@ -1309,6 +1309,19 @@ describe('Tool Registry', () => {
     });
   });
 
+  describe('complete_task review fields (#1314)', () => {
+    it('sends verdict, feedback and checks inside result', async () => {
+      mockClient.post.mockResolvedValue({ success: true, data: {}, status: 200 });
+      mockClient.get.mockResolvedValue({ success: true, data: [], status: 200 });
+      const checks = [{ entry: 0, did: 'opened it', outcome: 'pass' }];
+      await (tools.complete_task as any).execute({ workItemId: 'wi-v', sessionName: 'tl', summary: 'ok', verdict: 'verified', checks });
+      expect(mockClient.post).toHaveBeenCalledWith('/task-pool/complete/wi-v', {
+        agentId: 'tl',
+        result: { summary: 'ok', verdict: 'verified', checks },
+      });
+    });
+  });
+
   describe('report_status', () => {
     it('should send status via chat API with formatted message', async () => {
       mockClient.post.mockResolvedValue({
@@ -1376,6 +1389,19 @@ describe('Tool Registry', () => {
       );
     });
 
+    it('returns a refused completion instead of swallowing it (#1312)', async () => {
+      mockClient.post.mockImplementation(async (endpoint: string) => endpoint.startsWith('/task-pool/complete/')
+        ? { success: false, error: 'complete requires evidence', status: 400 }
+        : { success: true, data: { acknowledged: true }, status: 200 });
+      mockClient.get.mockResolvedValueOnce({ success: true, data: { workItems: [{ id: 'wi-running-3' }] }, status: 200 });
+
+      const result = await (tools.report_status as any).execute({ status: 'done', summary: 'Done' });
+
+      expect(result.acknowledged).toBe(true);
+      expect(result.completionError).toContain('wi-running-3 is still OPEN');
+      expect(result.completionError).toContain('complete requires evidence');
+    });
+
     it('should not auto-complete tasks when status is in_progress', async () => {
       mockClient.post.mockResolvedValue({ success: true, data: { acknowledged: true }, status: 200 });
 

package/packages/crewly-agent/src/runtime/tool-registry.ts

@@ -38,6 +38,15 @@ const completionEvidenceSchema = z.array(z.discriminatedUnion('type', [
   'if you could not finish (the WorkItem is then recorded as blocked, not done).',
 );
 
+/** What a reviewer checked when accepting a `Verify: …` item (#1314). */
+const reviewChecksSchema = z.array(z.object({
+  entry: z.number().int().nullable().describe("Index in the worker's output.evidence, or null for a direct look at the work"),
+  did: z.string().min(1).describe('What you did and saw: opened it, re-ran it (exit code), looked at the page'),
+  outcome: z.enum(['pass', 'fail']),
+})).describe(
+  'Required to accept a review: one entry per worker evidence entry you checked; at least one must name an entry when there is any.',
+);
+
 /** One completion evidence entry, as accepted by {@link completionEvidenceSchema}. */
 type CompletionEvidenceInput = z.infer<typeof completionEvidenceSchema>;
 
@@ -1286,8 +1295,11 @@ export function createTools(client: CrewlyApiClient, sessionName: string, projec
         sessionName: z.string().describe('Agent session that completed it'),
         summary: z.string().describe('Completion summary'),
         evidence: completionEvidenceSchema.optional(),
+        verdict: z.enum(['verified', 'rejected']).optional().describe('Reviewing a Verify item: verified accepts (needs checks), rejected sends it back (needs feedback)'),
+        feedback: z.string().optional().describe('With verdict rejected: what is wrong'),
+        checks: reviewChecksSchema.optional(),
       }),
-      execute: async ({ workItemId, sessionName: agent, summary, evidence }) => {
+      execute: async ({ workItemId, sessionName: agent, summary, evidence, verdict, feedback, checks }) => {
         // V3-only as of spec 2026-05-06-task-management-v1-deprecation.md.
         // Replaces v1 `/task-management/complete`. Note: the legacy
         // `absoluteTaskPath` parameter is removed; callers must now pass
@@ -1303,7 +1315,12 @@ export function createTools(client: CrewlyApiClient, sessionName: string, projec
         //   - tool-registry.ts §report_status auto-complete path (below)
         const result = await client.post(`/task-pool/complete/${workItemId}`, {
           agentId: agent,
-          result: completionResult(summary as string, evidence as CompletionEvidenceInput | undefined),
+          result: {
+            ...completionResult(summary as string, evidence as CompletionEvidenceInput | undefined),
+            ...(verdict ? { verdict } : {}),
+            ...(typeof feedback === 'string' && feedback.trim() ? { feedback } : {}),
+            ...(Array.isArray(checks) && checks.length > 0 ? { checks } : {}),
+          },
         });
 
         // Auto-cancel scheduled checks targeting the completing agent
@@ -1737,6 +1754,7 @@ export function createTools(client: CrewlyApiClient, sessionName: string, projec
         // V3-only as of spec 2026-05-06-task-management-v1-deprecation.md:
         // resolve the running claim from the pool, then call
         // `/task-pool/complete/:id`. Replaces v1 `/task-management/complete-by-session`.
+        let completionError: string | undefined;
         if (status === 'done') {
           try {
             const poolResp = await client.get(
@@ -1754,12 +1772,18 @@ export function createTools(client: CrewlyApiClient, sessionName: string, projec
               // `agentId` here is `sessionName` because that's the runtime
               // identity of the completing agent (the session whose
               // status=done message triggered this auto-complete path).
-              await client
+              // A refused completion (e.g. 400 evidence_required, #1312) is
+              // returned to the agent: swallowing it left the WorkItem open
+              // while the agent believed it was done.
+              const completion = await client
                 .post(`/task-pool/complete/${wiId}`, {
                   agentId: sessionName,
                   result: completionResult(summary as string, evidence as CompletionEvidenceInput | undefined),
                 })
-                .catch(() => {});
+                .catch((err: unknown) => ({ success: false, error: String(err) }));
+              if (!completion.success) {
+                completionError = `Status was sent, but WorkItem ${wiId} is still OPEN: ${completion.error ?? 'completion failed'}`;
+              }
             }
           } catch {
             // Non-fatal: status report still succeeded.
@@ -1777,6 +1801,9 @@ export function createTools(client: CrewlyApiClient, sessionName: string, projec
           }).catch(() => {});
         }
 
+        if (completionError) {
+          return { ...(result.success ? (result.data as Record<string, unknown>) : { error: result.error }), completionError };
+        }
         return result.success ? result.data : { error: result.error };
       },
     },