Multiple paying users have reported that Claude Code quotas were being consumed unreasonably. Previously, to update the terminal sub-agent status line, the official implementation required the sub-agent to fully reread the entire conversation context every 30 seconds, resulting in substantial token waste. A later optimization switched to using Haiku 5.5 to generate the status line from brief notes, cutting the cost of a single status line update by 99%. However, overall session quota consumption dropped by only about 5%, and the issue affected only command-line interface users; the desktop client was unaffected. In addition, Claude Code had previously blocked a large number of legitimate game development requests because the anti-plagiarism safety rules in the system prompt were too strict. Developers were flagged as infringing and denied answers merely for mentioning the names of well-known games to describe gameplay mechanics, and the official team later released a patch to fix this. Many users believe it is inappropriate for the official team to highly publicize a major bug that should have been fixed silently as an optimization achievement, and they demand that the platform reissue or reset quotas for users whose quotas were consumed for no reason. Some also remarked that some AI tools have flashy surface-level concepts while their underlying foundational engineering implementation is rough.