Skip to content

Security: palios-taey/claude-code-api-watchdog

Security

SECURITY.md

Security Policy

Scope

claude-code-api-watchdog is a local automation tool. It:

  • Reads your own tmux panes (tmux capture-pane)
  • Sends keystrokes to your own tmux sessions (tmux send-keys)
  • Optionally runs a relaunch command and an escalation command you configure

It does not make network calls, store credentials, or transmit any data off the machine. The only external commands it runs are the ones you pass via --resume-cmd and --escalate-cmd.

Things to know before you run it

  • Auto-restart of dead sessions is OFF by default. --resume-cmd is empty by default → if a Claude process dies, the watchdog escalates and does NOT relaunch. Opt in to auto-restart by setting --resume-cmd explicitly (e.g. claude --resume latest --dangerously-skip-permissions). The opt-in flag bypasses Claude Code's permission prompts on the relaunched session — that is appropriate for unattended automation but is a deliberate trust decision.
  • Use --no-restart for any session where re-running the last action on resume would be harmful (e.g. anything that posts, sends, or pays). The watchdog will nudge such sessions with Continue but never relaunch them.
  • The watchdog injects Continue into whatever session you name. Point it only at sessions you control.

Reporting a vulnerability

If you find a security issue in the watchdog itself (e.g. a way for pane content to trigger an unintended command), please open an issue without exploit details and request a private channel, or email the maintainer listed in the repo.

There aren't any published security advisories