fix(codex): preserve native safety behavior (#1024)

* test: reproduce issue #1020 * fix: preserve Codex safety behavior (closes #1020) * test: cover Codex safety buffering dismissal * test: cover Codex safety retry edge cases * fix: keep dismissed safety buffering prompts hidden
2026-10-15 20:29:55 +00:00 · 2026-07-12 18:39:56 +08:00
parent d160203bb2
commit d9ea9db507
8 changed files with 1166 additions and 14 deletions
@@ -72,7 +72,12 @@ describe('AppServerEventConverter', () => {
        expect(interrupted).toEqual([{ type: 'turn_aborted', turn_id: 'turn-1' }]);

        const failed = converter.handleNotification('turn/completed', { turn: { id: 'turn-1' }, status: 'Failed', message: 'boom' });
-        expect(failed).toEqual([{ type: 'task_failed', turn_id: 'turn-1', error: 'boom' }]);
+        expect(failed).toEqual([{
+            type: 'task_failed',
+            turn_id: 'turn-1',
+            terminal_source: 'turn_completed',
+            error: 'boom'
+        }]);
    });

    it('accumulates agent message deltas', () => {
@@ -693,6 +698,120 @@ describe('AppServerEventConverter', () => {
        expect(events).toEqual([{ type: 'task_failed', error: 'fatal' }]);
    });

+    it('preserves typed non-retryable cyber-policy errors', () => {
+        const converter = new AppServerEventConverter();
+
+        const events = converter.handleNotification('error', {
+            threadId: 'thread-1',
+            turnId: 'turn-1',
+            error: {
+                message: 'This content was flagged for possible cybersecurity risk.',
+                codexErrorInfo: 'cyberPolicy'
+            },
+            willRetry: false
+        });
+
+        expect(events).toEqual([{
+            type: 'task_failed',
+            thread_id: 'thread-1',
+            turn_id: 'turn-1',
+            terminal_source: 'error',
+            retryable: false,
+            codex_error_info: 'cyberPolicy',
+            error: 'This content was flagged for possible cybersecurity risk.'
+        }]);
+    });
+
+    it('preserves cyber-policy metadata from wrapped and completed-turn errors', () => {
+        const converter = new AppServerEventConverter();
+
+        expect(converter.handleNotification('codex/event/error', {
+            msg: {
+                type: 'error',
+                thread_id: 'thread-1',
+                turn_id: 'turn-1',
+                message: 'wrapped policy failure',
+                codex_error_info: 'cyber_policy',
+                will_retry: false
+            }
+        })).toEqual([{
+            type: 'task_failed',
+            thread_id: 'thread-1',
+            turn_id: 'turn-1',
+            retryable: false,
+            codex_error_info: 'cyber_policy',
+            error: 'wrapped policy failure'
+        }]);
+
+        expect(converter.handleNotification('turn/completed', {
+            threadId: 'thread-1',
+            turn: {
+                id: 'turn-1',
+                status: 'failed',
+                error: {
+                    message: 'completed policy failure',
+                    codexErrorInfo: 'CyberPolicy'
+                }
+            }
+        })).toEqual([{
+            type: 'task_failed',
+            thread_id: 'thread-1',
+            turn_id: 'turn-1',
+            terminal_source: 'turn_completed',
+            codex_error_info: 'CyberPolicy',
+            error: 'completed policy failure'
+        }]);
+    });
+
+    it('maps Codex model safety notifications', () => {
+        const converter = new AppServerEventConverter();
+
+        expect(converter.handleNotification('model/safetyBuffering/updated', {
+            threadId: 'thread-1',
+            turnId: 'turn-1',
+            model: 'gpt-5.4',
+            useCases: ['cyber'],
+            reasons: ['review'],
+            showBufferingUi: true,
+            fasterModel: 'gpt-5.4-mini'
+        })).toEqual([{
+            type: 'model_safety_buffering',
+            thread_id: 'thread-1',
+            turn_id: 'turn-1',
+            model: 'gpt-5.4',
+            use_cases: ['cyber'],
+            reasons: ['review'],
+            show_buffering_ui: true,
+            faster_model: 'gpt-5.4-mini'
+        }]);
+
+        expect(converter.handleNotification('model/rerouted', {
+            threadId: 'thread-1',
+            turnId: 'turn-1',
+            fromModel: 'gpt-5.4',
+            toModel: 'gpt-5.4-codex',
+            reason: 'highRiskCyberActivity'
+        })).toEqual([{
+            type: 'model_rerouted',
+            thread_id: 'thread-1',
+            turn_id: 'turn-1',
+            from_model: 'gpt-5.4',
+            to_model: 'gpt-5.4-codex',
+            reason: 'highRiskCyberActivity'
+        }]);
+
+        expect(converter.handleNotification('model/verification', {
+            threadId: 'thread-1',
+            turnId: 'turn-1',
+            verifications: ['trustedAccessForCyber']
+        })).toEqual([{
+            type: 'model_verification',
+            thread_id: 'thread-1',
+            turn_id: 'turn-1',
+            verifications: ['trustedAccessForCyber']
+        }]);
+    });
+
    it('maps thread/compacted notifications', () => {
        const converter = new AppServerEventConverter();
        const events = converter.handleNotification('thread/compacted', {