Only the latest release is actively maintained and receives security fixes. There is no long-term-support branch — fixes ship in the next tagged release.

Releases are checksummed and GPG-signed; see VERIFICATION.md for how to verify a download.

Please do not open a public GitHub issue for security vulnerabilities.

Use GitHub's private Report a vulnerability form instead.

Include:

• A description of the vulnerability and its potential impact
• Steps to reproduce or a proof-of-concept
• Affected version(s)

You can expect an acknowledgement within 72 hours and a fix or mitigation plan within 14 days for confirmed issues.

Areas most relevant to security review, with where they live:

Secrets

• Claude API key — AES-256-GCM in an encrypted file, or the system keyring when keyring is installed; key derived from the machine ID via PBKDF2-HMAC-SHA256 (gui/secret_store.py). Never written to the JSON config.
• NexusMods API key and SSO token — XOR+base64 obfuscation in the JSON config (gui/app_settings.py, _obfuscate/_deobfuscate). This is obfuscation, not encryption, and is documented as such — it stops casual shoulder-reading of a config file, nothing more.
• MCP server authorization tokens — same obfuscation, applied per entry (_obfuscate_mcp_entry), because the servers list is nested JSON.

Network and third parties

• NexusMods free-user download — reads session cookies out of Firefox/Chromium SQLite databases on the local machine (gui/nexusmods_client.py, _read_sqlite_cookies / find_browser_cookies).
• NexusMods SSO — a hand-rolled RFC-6455 WebSocket client over stdlib socket/ssl to wss://sso.nexusmods.com (gui/nexusmods_sso.py); frame encode/decode is the parsing surface.
• Claude MCP connector — when MCP servers are configured, chat content is sent to Anthropic, which connects to those servers and runs their tools server-side (gui/claude_client.py). Chat only: the translation pipeline never issues tool calls.
• Claude Code CLI backend — spawns the local claude binary as a subprocess (gui/claude_code_client.py). It strips ANTHROPIC_API_KEY and ANTHROPIC_AUTH_TOKEN from the child environment so the CLI cannot silently bill an API key, and runs in an empty scratch directory so no project CLAUDE.md is auto-discovered.
• Update check — polls the GitHub releases API and downloads the chosen asset over HTTPS to ~/Downloads, then opens the folder (gui/updater.py, gui/update_dialog.py). It never extracts or executes the download, and it does not verify the checksum — that is left to the user.

Parsing untrusted files (all of it runs on files downloaded from mod sites)

• Binary .strings/.dlstrings/.ilstrings (bethesda_strings/core.py)
• BA2 archives (bethesda_strings/ba2_handler.py)
• ESP/ESM/ESL plugins, including the VMAD script-property parser (bethesda_strings/esp_handler.py, vmad_handler.py)
• Scaleform SWF fonts and UI widgets (bethesda_strings/swf.py, font_checker.py, swf_widgets.py)
• ZIP / 7z / RAR extraction of downloaded mod archives (gui/nexusmods_browser_dialog.py, _extract_from_archive) — including the CLI fallbacks (7z, unrar) used when py7zr/rarfile are absent

Writing outside the project

• Plugin writers — EspFile.save() writes a translated copy to a path the user chooses, while apply_vmad_translations() rewrites the plugin in place (writing a .bak first). Both recompute record and GRUP sizes, so a parsing mistake becomes a malformed file the game loads.
• Desktop integration — gui/file_associations.py writes to $XDG_DATA_HOME on Linux and HKCU\Software\Classes on Windows, and registers a command line that Explorer/the file manager will execute with a user-supplied path.
• Audit log — append-only JSON-lines log of file operations and translation batches; it must never record string content (gui/audit_log.py).

• Vulnerabilities in third-party dependencies (report to the respective upstream project)
• Issues requiring physical access to the machine
• Social engineering
• Translation quality, model output, or anything an AI backend returns — incorrect or unsafe content is a bug report, not a security report
• The obfuscation of the NexusMods key being reversible: it is documented as obfuscation, and recovering it already requires read access to the user's config file