According to the comment, OpenAI Codex and the Claude systems often act only as terminals, so if a developer server experiences a memory leak that these models initiated, the memory usage will be incorrectly attributed to them. This can lead to misleading debugging and wrongful accountability; the example notes that a Ghostty process showing 100 GB of RAM can produce false conclusions.
Codex and Claude as terminals: risk of misattributing memory errors
According to the comment, OpenAI Codex and the Claude systems often act only as terminals, so if a developer server experiences a memory leak that these models initiated, the memory usage will be incorrectly attributed to them.
