[Bug] Claude model performance regression: instruction following and task completion degradation

Bug Description
intelligence regression again. same pattern as before. not following instructions, poor memory recall, cutting corners, missing things, takes about 5 - 7 iterations to resolve issues and really have to push the model hard to do work.

Environment Info

• Platform: darwin
• Terminal: vscode
• Version: 2.1.119
• Feedback ID: cd260244-e32d-4116-abfa-439b358c5d3e

Errors

[{"error":"Error: 1: Command failed with ERR_STREAM_PREMATURE_CLOSE: code --force --install-extension anthropic.claude-code\nPremature close \n    at kpL (/$bunfs/root/src/entrypoints/cli.js:1687:4046)\n    at async IpL (/$bunfs/root/src/entrypoints/cli.js:1687:1435)\n    at processTicksAndRejections (native:7:39)","timestamp":"2026-04-25T10:54:58.932Z"},{"error":"McpToolCallError: Could not connect to Chrome. Check if Chrome is running.\nCause: Failed to fetch browser webSocket URL from http://127.0.0.1:9333/json/version: fetch failed\n    at lb9 (/$bunfs/root/src/entrypoints/cli.js:4829:63)\n    at processTicksAndRejections (native:7:39)","timestamp":"2026-04-25T12:04:16.408Z"},{"error":"Error: Request was aborted.\n    at makeRequest (/$bunfs/root/src/entrypoints/cli.js:50:3448)\n    at processTicksAndRejections (native:7:39)","timestamp":"2026-04-25T14:09:28.867Z"}]