OpenSSF/OSV advisory MAL-2026-13428 confirms this npm version as malicious. The `conductor daemon` command (bin/conductor-daemon.js -> src/daemon.js) opens a WebSocket to https://conductor.conductor-ai.top (or a user-configured backend_url) and registers a message handler that, in `handleTerminalInput`, writes `payload.data` received from the server directly into a node-pty process. The PTY is spawned by `handleCreatePtyTask` running either the user's default shell (/bin/bash, /bin/zsh,...
Source
OpenSSF Malicious Packages via OSV
Summary
Malicious code in @love-moon/conductor-cli (npm)
Details
The `conductor daemon` command (bin/conductor-daemon.js -> src/daemon.js) opens a WebSocket to https://conductor.conductor-ai.top (or a user-configured backend_url) and registers a message handler that, in `handleTerminalInput`, writes `payload.data` received from the server directly into a node-pty process. The PTY is spawned by `handleCreatePtyTask` running either the user's default shell (/bin/bash, /bin/zsh, cmd.exe, $SHELL) or a CLI from DEFAULT_CLI_LIST, which launches `codex --dangerously-bypass-approvals-and-sandbox` and `claude --dangerously-skip-permissions` — the same flags are written into the generated config.yaml by bin/conductor-config.js. Any party controlling the WebSocket peer (the Conductor backend, an impersonator on the network, or a compromised backend account) can drive arbitrary commands into the installer's shell or into an AI CLI whose local approval/sandbox gates have been disabled by default. The remote-command channel and the disabled-safety flags together provide full-host command execution driven by conductor-ai.top.
Decision reason
OpenSSF Malicious Packages via OSV confirms @love-moon/conductor-cli@0.7.6 as malicious (MAL-2026-13428): Malicious code in @love-moon/conductor-cli (npm)