Planning Reliability, Prompt Transparency, and Admin Visibility

Retry flows are more durable, system prompts are now visible from the Agent tab, and super admins can manage agents from new global pages.

This release focused on making daily agent operations more dependable and easier to inspect. Planning runs now recover more cleanly, prompt behavior is more transparent, and admin navigation is stronger.

More Durable Planning And Retry Flows

Planning execution now handles retry chains and recovery scenarios with fewer edge-case failures.

  • Retry lineage is preserved so follow-up runs stay connected to their earlier attempts
  • Parent-child run relationships are clearer during ORCHESTRATE cycles, making run history easier to reason about
  • Critical path safeguards were tightened around execution start, recurrence behavior, and completion transitions to reduce stuck or inconsistent states

The result is a steadier execution loop when plans fail, retry, or recover mid-flight.

System Prompt Visibility In The Agent Tab

You can now open the full assembled system prompt directly from an agent profile and copy it when needed.

  • New prompt viewer dialog in the Agent tab
  • Copy-to-clipboard support for faster debugging and iteration

Alongside this, prompt assembly and defaults were streamlined so prompt behavior is more consistent and easier to maintain.

New Global Admin Views For Agents

Super admins now have a dedicated global agent area in Admin.

  • Agents index page to browse all agents across accounts
  • Agent detail page with account context and key usage stats
  • Navigation polish in admin so moving between operational views is faster

UI Polish Across Core Surfaces

Small but meaningful UX improvements landed across the app:

  • Live relative timestamps now refresh consistently in key views
  • Thinking-state chat feedback is clearer during response generation