취약점

-= Per source details. Do not edit below this line.=-

Source: amazon-inspector (0813d6ca6de1573ab8f99aae08444e589f4c5751931e4b18812140f720b74239)

Package self-describes as a 'Reverse-engineered Anthropic Claude Code CLI' and impersonates the legitimate @anthropic-ai/claude-code (bin name tme-claude, replicated slash commands, OAuth client IDs, and system prompts). When the user runs the CLI, authentication is hardcoded against https://tmeaicoding.tmeoa.com/token (dist/chunk-tvynx1j7.js line 28) and the resulting token is stored as process.env.TME_ANTHROPIC_API_KEY. All prompts, source code excerpts, and other developer-supplied input the user enters into the CLI are relayed through this publisher-controlled endpoint rather than to Anthropic. The English package description does not disclose the alternate destination. Additionally, the bundled auto-updater runs npm install -g @tmecontinue/claude@<version> --registry=https://mirrors.tencent.com/npm (dist/chunk-ass8j5mj.js line 5791), giving the publisher a persistent global-install code-update channel onto the installer's machine. No install-time lifecycle hooks fire on npm install; the harm fires when the operator runs the tme-claude CLI, but the silent-relay of caller-supplied prompt data to a publisher-controlled destination — combined with the impersonation of Anthropic Claude Code — is a supply-chain attack against developers who believed they were running the real Claude Code.