fix(cli): forward Claude API error records with null service_tier (#1857) (#2087)

Claude Code writes failed API calls ("Prompt is too long", usage limits,
logged out) as synthetic assistant records whose usage carries
`service_tier: null`. The schema only accepted undefined, so the local
session scanner dropped every one and the web session just looked stalled.

- Accept null for the usage fields the Messages API types as nullable.
- Keep the top-level `error`/`isApiErrorMessage` on assistant records in
  local mode, and `error` in the SDK converter for remote mode.
- Render assistant records carrying `error` as the existing error event,
  so the all-zero synthetic usage no longer resets the context gauge.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

web/src/chat/normalizeAgent.error.test.ts

@@ -29,4 +29,41 @@ describe('normalizeAgentRecord — agent error payloads', () => {
             }
         })
     })
+
+    it('normalizes Claude API error assistant records into error events without usage (#1857)', () => {
+        const normalized = normalizeAgentRecord(
+            'msg-2',
+            null,
+            1_700_000_000_000,
+            {
+                type: 'output',
+                data: {
+                    type: 'assistant',
+                    uuid: 'a1',
+                    isSidechain: false,
+                    isApiErrorMessage: true,
+                    error: 'invalid_request',
+                    message: {
+                        model: '<synthetic>',
+                        role: 'assistant',
+                        content: [{ type: 'text', text: 'Prompt is too long' }],
+                        usage: { input_tokens: 0, output_tokens: 0, service_tier: null }
+                    }
+                }
+            }
+        )
+
+        expect(normalized).toEqual({
+            id: 'msg-2',
+            localId: null,
+            createdAt: 1_700_000_000_000,
+            role: 'event',
+            isSidechain: false,
+            parentToolUseId: null,
+            content: {
+                type: 'error',
+                message: 'Prompt is too long'
+            }
+        })
+    })
 })

web/src/chat/normalizeAgent.ts

@@ -733,6 +733,29 @@ export function normalizeAgentRecord(
         if (data.isCompactSummary) return null
         if (!isClaudeChatVisibleMessage({ type: data.type, subtype: data.subtype })) return null
 
+        // Claude Code reports a failed API call ("Prompt is too long", usage
+        // limits, logged out) as a synthetic assistant message with a top-level
+        // `error`. Render it as an error, and keep its all-zero usage out of
+        // the context gauge.
+        if (data.type === 'assistant' && typeof data.error === 'string') {
+            const message = isObject(data.message) ? data.message : null
+            const text = Array.isArray(message?.content)
+                ? message.content
+                    .flatMap((block) => isObject(block) && typeof block.text === 'string' ? [block.text] : [])
+                    .join('\n')
+                : asString(message?.content)
+            return {
+                id: messageId,
+                localId,
+                createdAt,
+                role: 'event',
+                content: { type: 'error', message: text?.trim() || data.error },
+                isSidechain: Boolean(data.isSidechain),
+                parentToolUseId: asString(data.parentToolUseId) ?? null,
+                meta
+            }
+        }
+
         if (data.type === 'assistant') {
             return normalizeAssistantOutput(messageId, localId, createdAt, data, meta)
         }