Only the versions listed below receive security updates.

• Active - current release line. Receives new features, bug fixes, and security patches.
• Maintenance - previous release line. Receives security patches and critical bug fixes only; no new features.
• End of life - no longer supported. No fixes of any kind; please upgrade.

• Database at rest: the database is encrypted with AES-256-GCM (authenticated encryption). The key is derived from your password with Argon2id, with parameters stored per database (defaults: 4 iterations, 128 MiB memory, parallelism 4). The parameters are bounds-checked on load and out-of-range values are rejected (iterations 1 to 64, memory 8 MiB to 1 GiB, parallelism 1 to 16). Files use the v3 on-disk format: a portable big-endian header (magic, version, IV, salt, and the Argon2id parameters) that is fully covered by the GCM tag, so tampering with any header field is detected. v1 and v2 databases are read transparently and upgraded to v3 on the next save.
• Secrets in memory: derived keys, the decrypted token JSON, and per-token secrets live in libgcrypt secure memory. Those pages are mlock'd, so they are not paged to swap or written to a hibernation image, and they are wiped with explicit_bzero before being freed. On lock (manual, idle auto-lock, screensaver, or system suspend via logind) the master key and the decrypted database are purged from memory; unlocking re-derives the key rather than comparing a resident copy. Sensitive dialogs are closed and their fields, QR textures, and snapshots are cleared; pending dialog operations are canceled.
• Process hardening: PR_SET_DUMPABLE=0 and RLIMIT_CORE=0 are set at startup so a crash will not produce a core file containing secrets. Builds enable a stack protector, _FORTIFY_SOURCE=3, full RELRO with BIND_NOW, and PIE, plus optional control-flow protection, register clearing, strict flexible-array bounds, and automatic variable initialization when the toolchain supports them.
• Clipboard: copied OTPs are wiped after a configurable timeout, on database lock, and on exit (including SIGINT, SIGTERM, and SIGHUP), provided OTPClient still owns the clipboard. Another application's clipboard content is left intact.
• Exports: newly created encrypted migration exports require a nonempty password.
• Search provider: desktop search requires Secret Service access, a saved database password, an enabled provider, and a nonempty trigger keyword. Settings changes take effect live and revoke cached data and pending activation IDs. The keyword is a query filter, not authentication. Activation IDs are single-use, random 128-bit capability tokens with a 30-second TTL; OTP delivery is rate-limited through a single global bucket; and the derived key and caches are wiped after a period of inactivity.
• Out of scope: a same-UID attacker with ptrace or /proc/PID/mem access can read live secrets while the database is unlocked (distro-default kernel.yama.ptrace_scope=1 mitigates this for unrelated processes), and a cold-boot or DMA attack against a running, unlocked machine can recover secrets from RAM. Once the database is locked, the master key and decrypted database are no longer resident.

To report a vulnerability, email info@paolostivanin.com. Please do not open a public issue.

1. Acknowledgment - you will receive a reply within 24 hours confirming receipt and an initial assessment.
2. Fix - a patch will be developed and released within 7 days.
3. Disclosure - once the fix is published, a security advisory will be opened on GitHub.