| 1.0.537 | 2026-08-11 | - Snapshot-churn-risk verdict no longer reports "low" / "no material churn detected" when there was no prior validation to compare against at all; it now reports the assessment as unavailable.
|
| 1.0.536 | 2026-08-10 | - Write-action governance no longer reports "Safe to publish" or "No high-risk tools were detected" for a server whose write-action safety was never actually assessed; it now reports the assessment as unavailable.
- The executive verdict (Block/Allow decision) no longer recomputes its own independent blocker thresholds; it now derives from the same canonical production-readiness verdict every other surface uses.
- Comparison-template pages render the same canonical verdict label text as the server profile page, instead of an independently re-derived variant.
- Ownership/claim status is now derived from one shared function everywhere it is shown.
|
| 1.0.535 | 2026-08-10 | - Change detection (auth mode, write-action surface, tool snapshot) now compares against the most recent validation run that actually observed a tool surface, instead of the immediately-adjacent run -- a non-enumerating baseline no longer fabricates false high-severity alerts that degrade the production verdict.
- "Validation history" no longer shows a false 0 tool count for a run that never fetched a tool surface.
- Publishability status for Claude now weighs `admin_refresh_required` consistently instead of silently ignoring it.
- "Client compatibility" summary and gate-details checklist now derive from the same blocker list.
|
| 1.0.533 | 2026-08-10 | - A server no longer reads "Metadata only" after a real validation run was attempted and failed; a validated-and-failing server now reports a distinct "Failing" verdict.
- Publishability profiles now correctly report "blocked" when the underlying client-compatibility profile is blocked, instead of always downgrading to "caution".
- Six security-posture score components are now excluded from scoring, rather than credited a fixed partial score, when the underlying tool surface was never observed.
- Prompt Contract and Resource Contract score components now zero-anchor a `skipped` check status the same as `missing`, instead of crediting it as if it had run successfully.
|