Is there an existing issue for this?
Current behavior
When using Plane AI in Build mode (agentic work item creation), the agent
plans and executes its tool calls correctly (work items are created in the
backend). However, the UI does not transition to the final
"Awaiting response — Please confirm the actions you want to execute"
state on its own. Instead, the frontend remains stuck showing intermediate
reasoning steps ("Contemplating...", "Zeroing in on the best step...",
"Still working...") seemingly in an endless loop.
A manual page reload reveals that the backend had in fact already reached
the confirmation state. After the reload, the confirm step appears, can be
approved, and the actions complete successfully ("X successful actions").
So the agent terminates correctly — this appears to be a frontend state
sync issue, not an agent/looping problem.
Steps to reproduce
- Open Plane AI, switch to Build mode.
- Send a multi-step request that creates several work items, e.g.:
"Create a parent work item and 4 sub-tasks with priorities and short
descriptions for an employee onboarding process."
- Let the agent run. Observe the reasoning steps in the UI.
- Notice the UI appears to loop near completion and never shows the confirm step.
- Reload the page.
- Observe that the "Awaiting response — please confirm" state is now shown,
can be confirmed, and the actions complete.
Expected behavior
The UI should automatically transition to the "Awaiting response / confirm
actions" state once the backend reaches it, without requiring a manual reload.
Additional context
Backend tool calls all return HTTP 200 (verified in the LLM proxy logs), and
the created work items are correct. The only issue is that the frontend does
not surface the completed/awaiting-confirmation state until reloaded.
Environment
Production
Browser
Google Chrome
Variant
Self-hosted
Version
Commercial, v2.6.1 - Deployment: Docker (prime-cli managed) - LLM backend: custom (OpenAI-compatible proxy); reproducible across different models, so it is not model-specific.
Is there an existing issue for this?
Current behavior
When using Plane AI in Build mode (agentic work item creation), the agent
plans and executes its tool calls correctly (work items are created in the
backend). However, the UI does not transition to the final
"Awaiting response — Please confirm the actions you want to execute"
state on its own. Instead, the frontend remains stuck showing intermediate
reasoning steps ("Contemplating...", "Zeroing in on the best step...",
"Still working...") seemingly in an endless loop.
A manual page reload reveals that the backend had in fact already reached
the confirmation state. After the reload, the confirm step appears, can be
approved, and the actions complete successfully ("X successful actions").
So the agent terminates correctly — this appears to be a frontend state
sync issue, not an agent/looping problem.
Steps to reproduce
"Create a parent work item and 4 sub-tasks with priorities and short
descriptions for an employee onboarding process."
can be confirmed, and the actions complete.
Expected behavior
The UI should automatically transition to the "Awaiting response / confirm
actions" state once the backend reaches it, without requiring a manual reload.
Additional context
Backend tool calls all return HTTP 200 (verified in the LLM proxy logs), and
the created work items are correct. The only issue is that the frontend does
not surface the completed/awaiting-confirmation state until reloaded.
Environment
Production
Browser
Google Chrome
Variant
Self-hosted
Version
Commercial, v2.6.1 - Deployment: Docker (prime-cli managed) - LLM backend: custom (OpenAI-compatible proxy); reproducible across different models, so it is not model-specific.