[BUG] Tool response truncation occurs silently

[prestomation]

Problem Statement

Turns out some of my research tools have been unusable for my agent and I had no idea. I didn't discover this until I started dumping agent.messages to disk for review and noticed The tool result was too large! in replace of the tool response. Examining the code, it's not clear that even if I did enable debug logging that this behavior would be obvious to me during development. Even if it was, there is no mechanism today to handle this situation dynamically at runtime

Proposed Solution

I don't have a solution in mind but wanted to start with this issue for conversation

Use Case

I have research tools that browse the internet/read documentation pages/etc. The content is dynamic and not something I control and could be a large amount of content

Alternatives Solutions

As a workaround I can add pagination to my tool or use an llm to summarize the content.

Additional Context

No response

[zastrowm]

This is currently being done here:

sdk-python/src/strands/event_loop/message_processor.py

Line 159 in 8ffe24b

message["content"][i]["toolResult"]["content"] = [{"text": "The tool result was too large!"}]

We think this should be:

• Part of the context management
• Configurable

[Unshure]

Hi @prestomation , wanted to point you toward this issue: #152 (comment)

There was an infinite loop bug found in the implementation of the toolResult truncation logic. As a part of that fix, we plan to move this to the context manager and make it configurable!

[poshinchen]

Change has been merged, user will be able to use should_truncate_results=False to disable truncating the tool result in the next PyPI release.