Agent repeatedly interrupts workflow to reintroduce itself instead of continuing tasks
Problem
The Security Analyst Agent has a workflow interruption issue where it repeatedly stops its current task to reintroduce itself, preventing completion of requested work.
Steps to reproduce
- Start conversation with Security Analyst Agent
- Give initial prompt: "Tell me about yourself"
- Follow with task-oriented prompts like:
- "List and analyze vulnerabilities in this project"
- "Triage all vulnerabilities within the project and set those to the correct status with comments as you see fit"
- Observe agent behavior during task execution
Expected behavior
- Agent should complete requested tasks without interruption
- Agent should maintain context and continue working on assigned tasks
- Agent should only introduce itself when explicitly asked
Actual behavior
- Agent interrupts its workflow to reintroduce itself multiple times
- Agent stops mid-task and provides introduction instead of continuing work
- Agent appears to get stuck in a loop of self-introduction
- Even when prompted to continue with "Yes, please continue", agent responds with: "I appreciate you wanting to pause and understand who I am! Let me give you a clear picture of what I'm doing and who I am."
Specific examples observed
- Task interruption: Agent started vulnerability triage, dismissed some vulnerabilities, then stopped to reintroduce itself
- Repeated introductions: Agent kept trying to answer the initial "Tell me about yourself" question instead of proceeding with subsequent tasks
- Context confusion: Agent added notes to closed issues when asked to create progress tracking
Impact
This significantly impacts the usability of the Security Analyst Agent as it cannot complete complex multi-step tasks without getting stuck in introduction loops.
Technical notes
This appears to be a fundamental issue with Duo Chat conversation flow and context management rather than agent-specific logic.
Related
- Original feedback thread: #576916 (comment 2818913575)
- Session ID mentioned: 1171304
- Test project: https://gitlab.com/gl-demo-ultimate-fsieverding/devsecops-demo/devsecops-app
Screenshots
Multiple screenshots showing the agent repeatedly introducing itself instead of continuing with vulnerability triage tasks.
Edited by 🤖 GitLab Bot 🤖