claude-api

The document is a README describing an SDK that exposes powerful primitives: file read/write, shell execution, external MCP commands (runnable via npx), web/network access, environment variable control, hooks, and session persistence. The README itself is not malicious, but the documented features (notably 'bypassPermissions', 'acceptEdits', arbitrary MCP command execution, and hooks receiving file contents) represent high-risk functionality if present in the implementation and used without strict controls. I rate this as not directly malware but a moderate-to-high security risk surface that demands careful implementation, strong defaults, and user caution when enabling permissive modes or MCP servers.