[Bug] Claude Code generating incorrect solutions with high confidence in recent versions

Bug Description
Claude started making mistakes it did not make before. I used Claude Code over Codex specifically because it was much more reliable, but in the last few weeks I noticed that it makes mistakes. The tricky part is that these mistakes are very easy to spot and very easy to avoid making. So I am now doubting the reliability of the answers in areas that I cannot easily verify, or complex areas where mistakes are more likely to happen. And what's worst, is that it gives the wrong answer quite confidently....

Environment Info

• Platform: darwin
• Terminal: ghostty
• Version: 2.1.226
• Feedback ID: 89247446-4ce2-433f-8fd7-c16ecae8e181

Errors

[{"error":"Error: Streamable HTTP error: Error POSTing to endpoint: {\"id\":4,\"jsonrpc\":\"2.0\",\"result\":{\"content\":[{\"text\":\"Invalid JSON payload received. Unknown name \\\"endTime\\\": Cannot find field.\\nInvalid JSON payload received. Unknown name \\\"fullText\\\": Cannot find field.\\nInvalid JSON payload received. Unknown name \\\"startTime\\\": Cannot find field.\",\"type\":\"text\"}],\"isError\":true}}\r\n    at send (/$bunfs/root/src/entrypoints/cli.js:2702:6002)\n    at processTicksAndRejections (native:7:39)","timestamp":"2026-08-17T08:36:31.004Z"}]