Changelog

v0.34

Inspectible evidence and more resilient specialist scans

• Report findings now link directly to line-level evidence on GitHub, GitLab, and Bitbucket, while safely falling back to file-level references when exact lines are unavailable.
• Specialist scans now preserve useful natural-language research without depending on a brittle oversized output contract, and reports recover more cleanly from legacy or recap-wrapped specialist output.
• Records Desk scouting now runs through the normal ACP specialist flow, removing duplicate analyzer startup work while keeping live scan progress visible.
• ACP agents can now advertise and complete authentication flows, with clearer diagnostics and bounded recovery when credentials are missing or expire.
• The bundled ACP runtime now tracks an immutable Anvil master snapshot with Bifrost 0.8.21, and sniff-test quality, risk, and confidence scores are stored in queryable fields for more reliable benchmark and admin reporting.

v0.33

Resend email delivery

• Report notifications and operator email digests now send through Resend instead of SendGrid, using a Resend API key and verified sender domain.
• In-progress scans can accept report email signups again, and the site-wide banner once more advertises the free public beta.

v0.32

More resilient ACP scans and report delivery

• Configured ACP models are now validated before the backend accepts traffic, and specialist lanes retry with fallback models when a provider rejects or fails a request.
• ACP specialist outputs now use lane-specific contracts, so final synthesis can no longer duplicate or replace validated specialist artifacts.
• Scans with usable specialist evidence now retain a partial no-score report when final synthesis is unavailable instead of failing the entire case.
• Retryable synthesis partials are excluded from duplicate reuse and previous-report seeding so a later scan can produce a complete report.
• Report digests now deliver independently through email and Slack, with channel-specific retries and protected delivery-health visibility for operators.
• Slack report digests now keep operator links clickable without expanding them into link-preview cards.

v0.31

Scan-runner process cleanup

• The scan server now runs under a process reaper so completed Git helpers cannot exhaust its process and thread budget.

v0.30.1

Temporary email notification pause

• The site now clearly communicates an email-provider issue, and new report email registrations are temporarily paused while it is resolved.

v0.30

Current scan runtime and service notice

• Bundled scans now run Anvil 0.24.0 with Bifrost 0.8.12, using the latest published releases for the ACP executor and analyzer tooling.
• A site-wide notice now communicates that we are investigating recent scan failures.

v0.29

Current Anvil and Bifrost runtime

• Bundled scans now run Anvil 0.21.2 and Bifrost 0.8.0, keeping ACP execution and analyzer tooling aligned with their latest compatible published releases.
• The server image now builds with Rust 1.96 to support the upgraded analyzer runtime.

v0.28.2

Large repository memory guardrail

• The default degraded subset cap is now 10,000 files, reducing peak memory pressure on extremely large repositories.

v0.28.1

Large repository migration hotfix

• The production database migration for large-repository scan scope now applies as a forward-only schema change without modifying an already-applied baseline migration.

v0.28

Large repository scan safeguards

• Oversized repositories now continue in a degraded subset mode instead of exhausting analyzer memory after checkout.
• Scans that exceed the file limit clearly show that only a subset was analyzed and are kept out of public scoring.
• The default analysis cap is now 30,000 files, which protects production workers from repeatedly restarting on extremely large repositories.