Report privately through GitHub's Report a vulnerability form. Do not open a public issue for a security problem.

If GitHub Security Advisories are unavailable to you, open a normal issue that says only "security report, please provide a private channel" — without any detail — and a contact will be arranged from there.

What to expect:

This is a single-maintainer hobby project, not a funded product. Those windows are what one person can realistically hold to, not a service-level agreement. If a deadline slips you will be told, and told why.

The parts of this project that touch your machine:

• The exporter (Scripts/ccwidget-export.py.template) — it runs on every status line redraw with your privileges. Anything that lets it write outside its exchange directory, follow a symlink, execute injected code, or consume unbounded resources is in scope.
• The installer (App/Installer.swift, App/SettingsEditor.swift) — it edits ~/.claude/settings.json and writes an executable script. Anything that corrupts the config, escapes the intended paths, or installs without an explicit user action is in scope.
• The exchange format — anything in snapshot.json or history.jsonl that discloses more than intended, or that can crash or hang the widget when malformed.
• Log privacy — user paths, project names or raw field values reaching the unified log as .public.

• Attacks requiring an attacker who already runs code as you. Such an attacker can edit ~/.claude/settings.json directly. Report it anyway if this project amplifies that position — for example by turning it into a repeated write to files elsewhere — but plain "same-user process can do X" is not a finding here.
• The absence of Developer ID signing and notarization. Known, documented, and tracked separately.
• Anything about Claude Code or Anthropic's services. Report those to Anthropic. This project only reads a file Claude Code hands to it.

This project has no network code, no credentials, and no privileged helper. It reads a JSON blob that Claude Code pipes to a status line command, writes two files into the widget extension's own container, and renders them. The security-relevant surface is therefore file handling and config editing — not authentication, not transport, not cryptography. Findings that matter are the ones where a file gets written somewhere it should not, or where the config is left broken.