Skip to content

Security: ksudo-dev/core-terminal

SECURITY.md

Security policy

Supported versions

Version 0.2.2 is the supported release line. Older package files are not maintained.

Reporting a vulnerability

Use a private GitHub Security Advisory for this repository. Include the Core Terminal version, Ubuntu release, architecture, reproduction steps, expected behavior, observed behavior, and a log or backtrace with credentials and terminal input removed. Do not publish an unpatched vulnerability in an issue.

The project will acknowledge the report through the private advisory and post a fix or status update there. No fixed response time is promised.

Threat boundaries

Core Terminal starts the configured shell as the current user. A shell or terminal program can read and modify files and processes available to that user. Core Terminal does not sandbox shell commands, imported profile values, or programs started by the shell.

The Flatpak grants home-directory access and access to the Flatpak host-command interface. Core Terminal uses that interface only to start the configured host shell and its requested working directory. This permission is intentional: a sandbox-only shell could not run the user's installed commands. Programs started through the Flatpak have the same user privileges as native terminal programs. The bridge relies on the host-command broker's host-session environment and overrides only TERM, COLORTERM, VTE_VERSION, and an explicitly configured locale. It does not copy the Flatpak process environment over the broker-provided host environment.

The Flatpak host supervisor requires Linux 5.3 or newer and uses pidfds rather than numeric PIDs for cleanup. It cannot signal a process that has changed to a user or security domain the current user may not control. Such a privileged process can outlive its terminal session, so cleanup cannot be guaranteed. The helper exits with a failure status, but this release does not show a dedicated cleanup-error dialog.

JSON settings and profile documents are bounded to 4 MiB. .terminal imports are parsed as plist data; XML entity and unsupported document-type constructs are rejected. Imported values are mapped to profile fields and are never executed. The application validates terminal type, locale, colors, dimensions, and absolute shell paths before using them. Profile JSON does not store credentials.

The custom command option intentionally runs through the selected login shell. The direct command path uses GLib's argument parser, which preserves quotes and backslash escapes without shell expansion, substitution, or redirection. Invalid direct commands fall back to the configured login shell. Users must treat both paths as commands with the current user's privileges.

Custom key mappings are decoded into bounded PTY byte sequences. They cannot start a process by themselves, but the active shell or terminal program can interpret the bytes as input. Imported mapping entries pass the same decoder before they are used.

VTE interprets escape sequences emitted by the child process. Opening a URI or using a terminal program remains a user action and uses the desktop security boundary. Wayland does not provide a Core Terminal switch equivalent to macOS Secure Keyboard Entry.

Private inputs and package contents

profiles-private/, supplied screenshots, .terminal files, and reference archives are not runtime inputs. Package builds copy explicit source and file allowlists. scripts/check-private-data.sh checks publishable paths, package contents, and absolute developer home paths in package files. Run it before publication.

The release checklist also runs cargo audit, scans publishable files for credential patterns, checks the packaged ELF dependencies and dynamic section, and compares the Debian payload against its documented file set. Commands are listed in docs/TESTING.md.

There aren't any published security advisories