Feedback

Persona Description Does Not Clearly Influence Agent Behavior
Product feedback submitted from Reflex. Description : Steps to Reproduce Create a Session using an OpenCode agent. Perform a development/QA task in the Session. From the Session, select Save as Persona. Create a Persona named Python Development & QA Agent with a description stating that it is a reusable development and QA agent capable of implementation, verification, investigation, and providing test summaries. Save the Persona. Navigate to Agent Personas and launch the newly created Persona. Observe that the Launch screen displays the Persona description and the captured configuration. Replace the captured To-Do List prompt with a different task, such as creating a Python Calculator application. Launch the new Session and allow the agent to complete the task. Ask the agent: Based only on the instructions and configuration actually provided to you, what role, Persona-specific instructions, or specialization did you inherit from the Python Development & QA Agent Persona? Ask the agent to distinguish which behaviors came from the Persona, the underlying OpenCode configuration, and the task-specific prompt. Observe that the agent identifies its verification behavior as coming from the base OpenCode instructions and the written verification summary as coming from the task prompt, but cannot identify any Persona-specific Python/QA instructions. Actual Result The launched agent could not identify any Persona-specific instructions or specialization. It reported that its behavior was based on the underlying OpenCode configuration/default instructions and the current task prompt. Expected Result The launched agent should either: receive and be able to identify the behavioral instructions defined by the Persona, or Reflex should clearly communicate that the Persona description is informational/metadata and does not influence the agent's behavior. What Should Change Reflex should clearly distinguish between Persona metadata and instructions actually passed to the agent. If the Persona description is intended to define agent behavior, those instructions should be explicitly applied to the launched agent. Alternatively, the Persona UI should clearly show which configuration and instructions will affect the launched Session. Impact Users may assume that creating a specialized Persona such as Python Development & QA Agent will cause future Sessions to inherit that specialization. Without visibility into what is actually applied, users cannot confidently understand the behavior or value of the Persona. Filed from an agent session (Python Calculator Application Implementation Verification).
1
Automation flow is difficult to understand and create for users
Product feedback submitted from Reflex. Description : Issue: The current Automation Builder flow is not intuitive enough for a new or regular user. It is not immediately clear how the different nodes should be connected, what each node is supposed to do, or which node should be used for a particular use case. Creating an automation requires the user to understand the purpose and relationship between multiple components such as Trigger, Agent Task, Agent, Approval, Router, and Action before they can successfully build a flow. User Impact The overall flow is difficult to understand at first glance. Users may not know which node to select or where to start. The relationship between different nodes is not always obvious. Users may need prior knowledge of the platform before creating an automation. Building a simple automation can feel unnecessarily complex. There is limited guidance when configuring the flow and connecting nodes. Expected / Recommendation The Automation Builder should provide a more intuitive and guided flow-creation experience. Possible improvements include: Provide a guided workflow/template for common automation scenarios. Clearly explain the purpose of each node when adding it. Provide contextual help/tooltips for nodes and configuration fields. Show examples of how nodes should be connected. Provide suggested next nodes based on the current node. Clearly indicate where the flow should start and end. Provide simple starter templates such as: Trigger → Agent → Action Trigger → Agent Task → Approval → Router Trigger → Agent → PASS/FAIL branches Provide validation or guidance while building the flow instead of only showing errors after Create is clicked. QA Recommendation The Automation Builder could be made significantly easier to use by moving from a configuration-first experience toward a more guided, visual workflow-building experience, especially for first-time users.
0
Load More
→