Users report Claude coding-quality regression

Anthropic users said on May 22 that Claude’s coding quality and context retention had deteriorated again, reviving a complaint the company had already addressed in an April engineering post. Posts on X described weaker code output, lost context and behavior that users said also showed up in Claude-powered tools such as Slack and Linear. Anthropic has not published a new postmortem for the May complaints. Anthropic said in an April 23 engineering update that it had traced an earlier wave of Claude Code quality reports to three separate changes affecting Claude Code, the Claude Agent SDK and Claude Cowork. The company said the API and inference layer were not affected, and that the three issues had been resolved as of April 20 in Claude Code v2.1.116. ### What did Anthropic say caused the earlier degradation? Anthropic said on April 23 that one change was made on March 4, when Claude Code’s default reasoning effort was reduced from high to medium to cut latency. The company said it reversed that change on April 7 after users told it they preferred “higher intelligence” by default and would opt into lower effort for simpler tasks. (anthropic.com) Anthropic also said a March 26 change to clear older thinking from sessions idle for more than an hour introduced a bug that kept repeating every turn, making Claude “seem forgetful and repetitive.” The company said it fixed that bug on April 10. A third change, added on April 16 to reduce verbosity, hurt coding quality in combination with other prompt changes and was reverted on April 20, Anthropic said. (anthropic.com) ### Why are users talking about this again in late May? May 22 posts on X said the problems had not fully gone away for some users, with one user saying the decline had been visible for about a month and appeared across Claude integrations including Slack and Linear. Those posts match Anthropic’s own April account that reports had been coming in over “the past month,” though the company has not publicly tied the new May complaints to a fresh root cause. (anthropic.com) Anthropic’s April post said the earlier reports were hard to distinguish from normal variation at first and that internal usage and evaluations did not initially reproduce the issues users were describing. The company said that because different changes affected different slices of traffic on different schedules, the combined result looked like “broad, inconsistent degradation.” (anthropic.com) ### What changed in Claude Code this week? Anthropic’s Claude Code changelog shows version 2.1.147 was released on May 21. That release kept pinned background sessions alive when idle, renamed `/simplify` to `/code-review`, and added GitHub pull-request commenting through the code-review flow, according to the changelog and GitHub release notes. Anthropic released Claude Code v2.1.148 on May 22. (anthropic.com) The company said that patch fixed “the Bash tool returning exit code 127 on every command for some users,” calling it a regression introduced in v2.1.147. ### What are users doing as a workaround? Anthropic’s documentation still lists slash-command tools including `/code-review`, and users on X circulated advice to use `/compact` or external dashboards to watch for context degradation. (code.claude.com) Anthropic has not published official guidance in the sources reviewed here saying `/compact` is a remedy for the May complaints. Anthropic’s April 23 post said it was resetting usage limits for all subscribers as of that date and outlined process changes intended to reduce the chance of similar quality regressions. (code.claude.com) The next public checkpoints are likely to be the Claude Code changelog and any new Anthropic engineering update if the company decides the May complaints warrant another postmortem. (anthropic.com)