Skip to content

AuraScan v0.5.0

Choose a tag to compare

View all tags
@crizzler crizzler released this 15 Jul 06:02
· 22 commits to main since this release
v0.5.0
945f721

AuraScan v0.5.0

Assisted background recovery and AI-guided repair planning release. AuraScan's
AI can now choose useful local diagnostics and help prioritize a repair plan
without receiving command-generation or repair authority.

Highlights

  • Added a two-pass incident AI workflow. The triage pass may select only
    AuraScan-owned opaque diagnostic probe IDs; the final pass explains and ranks
    only locally verified repair actions.
  • Added bounded read-only probes for package integrity, kernel headers and
    DKMS, initramfs readiness, failed noncritical services, cache pressure,
    repository health, and stale package-manager locks.
  • Added strict execution limits: at most 24 probe candidates, six AI-selected
    probes, 12 executed probes, 180 seconds of local probe work, and two bounded
    provider requests per incident.
  • Added separately opted-in Assisted Background Recovery through a hardened
    systemd --user timer. It can prepare a private repair plan while the user is
    logged in, but cannot run sudo, invoke repairs, or write system paths.
  • Added six-hour prepared-plan reuse with evidence fingerprint matching. The
    foreground resolution flow refreshes probes and privileged preconditions
    before presenting one combined repair confirmation.
  • Added provider retry backoff, per-UID processing, private background state,
    and a concise tray notification when a verified plan is ready.
  • Added an offline root Safe Autopilot service, disabled by default. Its
    unattended allowlist remains limited to reversible stale pacman-lock recovery
    and verified mirrorlist restoration, with manifests, rollback, and a 24-hour
    identical-action cooldown.
  • Extended incident reports to schema 1.3 while retaining compatibility with
    1.0 through 1.2 reports.
  • Extended the setup wizard and Doctor for background-AI readiness, user timer
    state, probe tools, retry state, root repair policy, and automatic-repair
    outcomes.
  • Added a dedicated privacy document describing evidence bounds, redaction,
    storage, AI request limits, and the privilege boundary between collectors,
    the user assistant, and repair execution.

Enablement

Existing foreground incident diagnostics continue to work without background
processing. The new features are explicit opt-ins:

aurascan init --enable-incident-background-ai
aurascan incidents --background-ai-status
sudo aurascan incidents --auto-repair safe

Safe Autopilot may also be left at its default off setting while background
AI prepares broader repairs for foreground confirmation.

Safety Notes

Root boot and weekly collectors remain network-isolated and read-only. The
networked assistant runs unprivileged as the logged-in user. AI can select only
known probe and action IDs; it cannot provide executable targets, commands,
file edits, approval, or a successful repair result. AuraScan rebuilds every
privileged action from trusted current state and revalidates it as root.

Filesystem repair, partition and bootloader changes, authentication policy,
firmware operations, user-data deletion, AUR rebuilds, and automatic rebooting
remain prohibited.

Assets 2
Loading