kapi handles credentials, tokens and requests to private APIs, so security reports are very welcome.

Please don't open a public issue. Use GitHub's private vulnerability reporting instead, with steps to reproduce and the version you tested. You should hear back within a week.

Only the latest release gets fixes. kapi updates itself in place (Help → Check for Updates…).

• Collection scripts (pre-request and test) run in a sandbox: a disposable Web Worker with no Tauri IPC in the app, and a fresh node:vm context in the CLI and MCP server. They can't read files, run commands or reach the app's own state; their only way out is pm.sendRequest.
• The webview runs under a Content-Security-Policy, and can only call the native commands listed in src-tauri/capabilities/. git is limited to the exact invocations kapi makes.
• Secret variables are encrypted at rest (AES-256-GCM) with a key held in the OS keychain.
• AI CLIs receive prompts over stdin, never as command-line arguments.
• Updates are only installed when signed with the key whose public half ships in the app.