Skip to content

Security: wastedcode/claudemux

SECURITY.md

Security Policy

Supported versions

claudemux is pre-1.0. Security fixes land on the latest published minor; please upgrade to the most recent release before reporting.

Version Supported
latest 0.x
older 0.x ❌ (upgrade first)

Reporting a vulnerability

Do not open a public GitHub issue for a security vulnerability.

Report privately via one of:

Please include: affected version, a description of the issue, reproduction steps or a proof-of-concept, and the impact you foresee. We aim to acknowledge within 72 hours and to agree a disclosure timeline with you before any public write-up.

Threat model — what claudemux does and does not touch

Understanding the boundary helps you judge whether something is a claudemux vulnerability or expected behavior:

  • claudemux spawns and drives a real, already-authenticated claude CLI on a box you control. It inherits that box's claude configuration (auth tokens, permission mode, model, MCP) and owns no credentials of its own. It never reads or writes your claude auth; it has no API keys.
  • It shells out to tmux. Every invocation carries -f /dev/null so your ~/.tmux.conf is never read, and the public surface is backend-neutral. The Backend interface exposes no sendRawText primitive — input cannot leak around the seam.
  • It never matches a peer process by name. PID/PGID captured at spawn only, so it cannot accidentally kill another claude REPL running as the same user (see docs/decisions/0004).
  • Session names are validated (src/session/validate.ts) to keep control/reserved characters out of backend argv.
  • Workspace trust fails closed. Pointing a session at a never-trusted folder raises WorkspaceUntrusted unless you explicitly pass --trust-workspace, which writes a persistent per-folder authority grant — treat that flag as you would sudo for that directory.

Out of scope

  • The behavior, output, or safety of the underlying claude CLI and the agent it runs — report those to Anthropic.
  • Anything a consumer's own policy layer (a watchdog, an orchestrator) does on top of claudemux.
  • Running against code or a workspace you don't trust after deliberately passing --trust-workspace — that flag is the documented authority grant.

Thanks for helping keep claudemux and its users safe.

There aren't any published security advisories