ChatGPT Usage Monitor reads account-level limits through a locally installed Codex CLI or the backend bundled with a supported ChatGPT desktop package. The applet does not read credential files, accept API keys or persist usage responses. Please use a private channel for anything that could expose Codex or ChatGPT authentication data.

Security fixes target the latest published release and the current main branch. This policy also applies after 1.0: older releases do not receive separate backports; users should update to the latest supported release.

Use GitHub private vulnerability reporting when possible. If that channel is unavailable, contact the repository owner privately through the address published on the owner's GitHub profile.

Do not open a public issue containing credentials, tokens, Codex account files, private paths, complete analytics responses or screenshots with account-specific data. Include the smallest reproducible description, affected version, Linux Mint and Cinnamon versions, selected backend type/version, expected security boundary and a safe synthetic reproduction when possible.

• Authentication and network access remain the selected official backend's responsibility. The applet does not read its credential files.
• Normal refreshes send one read-only account/rateLimits/read request.
• Only explicit confirmation starts account/rateLimitResetCredit/consume. No startup, refresh, discovery or timer path redeems a reset.
• A timeout or lost reply leaves an unknown outcome. A private journal stores the idempotency key, selected opaque credit ID (if any) and backend path before dispatch. Retrying requires a new confirmation and reuses that same request. A recognized result clears the journal; corrupt journal data disables redemption. Keep the same account until the attempt is resolved.
• The normalized snapshot and available reset-credit details stay in memory. History stores only model/window keys, sample times, percentages, window reset timestamps and sampled numeric credit balances needed to derive observed credit consumption for eight days, capped at 10,000 samples. It never stores full provider responses, prompts or tokens.
• The 7d reset tooltip may retain a separate local observation journal with only model/window keys and reset timestamps. It is written with mode 0600 beside history and contains no provider response, prompt or credential data.
• History and the unresolved-reset journal live under $XDG_STATE_HOME/cinnamon-chatgpt-usage (normally ~/.local/state/...), in files created with mode 0600. Newly created state directories use 0700. Uninstall retains these files and settings. The journal persists until its attempt is resolved; it has no automatic expiry that could enable a second logical redemption.
• History is local to a desktop profile, not partitioned by account. Follow the documented tracking reset procedure before changing accounts. A backend path match alone cannot establish that the signed-in account is unchanged.
• Backend commands receive argument arrays, never shell-interpolated paths. Discovery and version probes run outside Cinnamon's UI thread. Each version probe has a two-second deadline and a bounded output buffer.
• JSON-lines transport limits response size and request time, and terminates and reaps its backend process group on completion or cancellation. Applet removal sends SIGTERM to the helper so cleanup can finish.
• Web shortcuts open fixed official ChatGPT URLs. No unrelated telemetry or network endpoint is used by this project.

The project does not defend against a malicious process already running as the same desktop user. Codex, ChatGPT, Cinnamon, Python and the operating system are external trust boundaries. Report an upstream vulnerability to its owner and also report it here privately if this integration makes the impact worse.