Terminal

Run arbitrary shell commands inside any sandbox.

Where to find it

Sandbox table → row → Terminal tab.

What it is

A thin wrapper around E2B's command execution. You type a command, the UI ships it to the sandbox, and streams stdout + stderr back. It's not an interactive PTY — think ssh user@host 'some command' rather than a full shell.

Commands you'll actually run

pm2 status                      # which processes are up?
pm2 restart shmastra            # restart the dev server
pm2 restart healer              # restart the healer agent
pm2 logs shmastra --lines 100   # tail server logs
ls .logs                        # where log files live
cat .env                        # inspect env (handle with care)
pnpm install                    # re-install after a messy merge
pnpm dry-run                    # run the Shmastra dry-run build manually
git status                      # see uncommitted changes
git log --oneline -n 20         # recent commits (includes healer's)
du -sh node_modules files .storage   # disk usage

Combining with other tabs

• Chat with agent is a friendlier way to do investigations — the Claude session narrates as it goes.
• Logs shows pm2 output live; use the terminal when you want to run things between log reads.
• Files is easier for edits than piping through sed.

When a sandbox is broken

A typical rescue loop:

1. Terminal: pm2 logs shmastra --lines 200 — capture the last error.
2. Files: find and fix the offending file.
3. Terminal: pm2 restart shmastra && sleep 5 && pm2 status.
4. Sandbox table: confirm status flips back to ready.

If it doesn't, hand off to chat: "Here's the error, have a go."

Be careful

• Destructive commands are unprotected. rm -rf works. Don't rm -rf /.
• Secrets are readable. cat .env shows real values; a shoulder surfer can see them on your screen.
• No history between tabs. Each terminal request is independent — environment variable exports don't carry across.