Agent responds with "Understood" without taking action as conversation context grows

[zcyee]

Before opening, please confirm:

• I have searched for duplicate or closed issues

Operating System

darwin

Kiro Version

0.11.131

Bug Description

As the conversation context grows longer, the agent increasingly responds with just "Understood" without actually performing any action or producing meaningful output. The longer the context becomes, the more frequently this occurs. Current workaround: I added a steering rule to have the agent proactively monitor context length. When the conversation becomes too long, the agent asks whether to switch to a new window and provides a recovery prompt containing the current task progress and a list of key files to read.

Steps to Reproduce

1. Open Kiro and start a conversation with the agent
2. Work through a series of tasks in the same conversation (e.g., multiple spec tasks, code generation, file edits)
3. As the conversation grows longer (typically after 15-20+ back-and-forth exchanges), send a new request to the agent
4. Observe the agent responds with just "Understood" without performing any action or producing meaningful output
5. The longer the context, the more frequently this happens — eventually almost every response becomes "Understood" with no actual work done

Expected Behavior

The agent should continue to process requests and produce meaningful output regardless of conversation length. If the context window is approaching its limit, the agent should either automatically compact the context and continue working, or proactively inform the user that a new conversation is needed — rather than silently responding with "Understood" and doing nothing.

Conversation ID

7d3a53d4-4361-4949-bd25-9b274fe4725f
7d3a53d4-4361-4949-bd25-9b274fe4725f
7d3a53d4-4361-4949-bd25-9b274fe4725f
7d3a53d4-4361-4949-bd25-9b274fe4725f
7d3a53d4-4361-4949-bd25-9b274fe4725f

Additional Context

No response

[github-actions]

🤖 Potential Duplicate Detected

This issue appears to be similar to:

• #6445: Terrible chat message context usage (85% similar)
• #6395: kiro stopped working or responding (85% similar)
• #6183: No response after gradle-related tool using. (1.27) (85% similar)
• #6132: Automatic Sprint/Task Review Before Context Compression (85% similar)
• #5527: Hooks bypass auto-compaction trigger - ValidationException at 100% context (85% similar)
• #5457: Context usage always shows 0% and crashes with "unexpected error" when context is full (85% similar)
• #5165: after conversation compating, Kiro undoes the previous changes (85% similar)
• #5046: AI Chat messages appear in history but are not processed — duplicates added on retry, requires multiple sends or long wait (85% similar)
• #4992: New sessions immediately hit 85%+ context usage and enter summarization loop, even with minimal steering docs and simple prompts. This worked fine for 6+ months prior. (85% similar)

What happens next?

• ⏰ This issue will be automatically closed in 3 days
• 🏷️ If this is not a duplicate, you can prevent automatic closure by adding a comment or reacting with 👎 to this message.
• 💬 Comment on the original issue if you have additional information

Why is this marked as duplicate?
Both report issues with chat context management - the new issue describes agent responding with 'Understood' without action as context grows, while #6445 reports 'terrible chat message context usage' requiring frequent new chats and context re-provision, indicating the same underlying context handling problem