Skip to main content

Release 0.5.22 — Simplified Admin, Smarter Access Control, and a Leaner Architecture

Sri Aradhyula
Sri Aradhyula

Released: 2026-06-25 Chart: oci://ghcr.io/cnoe-io/charts/ai-platform-engineering:0.5.22 Previous release: 0.5.21

Highlights​

0.5.22 is a substantial feature release. The legacy supervisor routing model and standalone A2A agents are removed — the platform now routes entirely through the dynamic agent layer. The admin experience is overhauled: a new team-based user access view replaces raw FGA tuple management, Webex space access is now correctly tied to agent assignment, and release notes move to a per-user preference. AI Review grading is more consistent, and CI receives two rounds of security hardening.

What's New​

Legacy supervisor and standalone agents removed​

The legacy supervisor model and A2A standalone agents have been removed from the platform. All routing now flows through the dynamic agents layer, which has been the default path for several releases. This simplifies the architecture and removes maintenance overhead for patterns that are no longer in active use. (#1728)

Team-based user access view in admin​

The admin panel replaces low-level OpenFGA tuple management with a team-centric view of user access. Admins can now see who has access to which resources through the lens of team membership — without needing to understand the underlying authorization model. (#2016)

Webex space access tied to agent assignment​

Webex space access is now granted based on which agent is assigned to a space, not just general team membership. This means spaces are only accessible through the agent that owns them, giving tighter isolation when multiple agents serve different Webex spaces. (#1763)

Release notes as a per-user setting​

Release notes notifications have moved from a global platform setting to a per-user preference under General settings. Users can opt in or out independently, and the settings page is simpler as a result. (#2017)

The Slack bot's footer now uses bold-italic formatting for its hint text, making the "reply to this thread" nudge more visually distinct from the main response. (#2011)

Bug Fixes​

  • AI Review grade not persisted on creation: When a review was created, the initial AI-assigned grade was not saved to the record, so it disappeared until the page was reloaded. The grade is now persisted on the create call. (#2014)
  • AI Suggest output misaligned with rubric: The AI suggestion in review mode was not constrained to the rubric criteria, sometimes returning suggestions that did not map to the review categories. This is now aligned. (#2014)
  • Blocking message UX in AI Review: The blocking message shown while an AI review is in progress was not dismissing correctly in some cases. Fixed. (#2014)
  • Slack channel diagnostics loaded all OpenFGA tuples: The diagnostics query was pulling all tuples from OpenFGA client-side and filtering locally, which caused slow loads and potential timeouts for large deployments. Filtering is now done server-side. (#2009)
  • CI workflow security findings: Resolved zizmor-flagged template injection risks, unpinned actions, and overly broad workflow permissions across two rounds of fixes. (#1998, #2010)

Breaking Changes​

The legacy supervisor model and standalone A2A agents are removed. If you have any values that reference those components, they can be dropped. Dynamic agents remain the supported routing layer.

Upgrade Guide: 0.5.21 → 0.5.22​

Helm Values Changes​

No required values changes. If your values.yaml references the legacy supervisor or standalone A2A agents, those entries can be removed — they have no effect from this version onward.

Upgrade Runbook​

helm upgrade ai-platform-engineering \
  oci://ghcr.io/cnoe-io/charts/ai-platform-engineering \
  --version 0.5.22 \
  -f your-values.yaml

Verify:

kubectl get pods -n <namespace>