Reports say Anthropic’s Claude Code feature that lets users “don’t ask again” can record shell commands exactly as approved, including any passwords or API keys typed into them. Multiple outlets describe cases where those stored commands end up in files that are present in public GitHub repositories, making the credentials readable to anyone.
The issue is tied to developers approving shell commands inside the coding assistant; one report says the assistant saves the approved command “word-for-word” into a project settings file when that option is selected. Security firm Apiiro is cited in at least one source as the entity identifying the scope, including the number of affected public copies.
Across coverage, Hacker News and Medium focus on the risk that credentials can be unintentionally persisted and then published through GitHub. The framing differs slightly in wording—some emphasize “don’t ask again,” while another describes “yes, and don’t ask again”—but all accounts point to the same underlying mechanism and security concern.
