SubAgents Fail Writing 25kb File - Sequential
Hey everyone, I'm able to reproduce a bug across my OpenClaw agents. I plan on making a bug report.
There's a known bug (#4355) regarding parallel agents, but this tests a bug with sequential agents that should avoid that bug.
It can be easily modified to request smaller file sizes (10kb, 15kb, etc) , which seem to lead to more successes. I get consistent failures at 25kb file requests.
If you don't mind, could you throw this prompt at your agent and see if it also fails?
....
Please perform only the following test.
Use sequential subagents only. Do not perform any part of this test yourself and do not call subagents in parallel.
If ANY subagent fails. IMMEDIATELY STOP THE TEST and report the results so far. Inform me which subagent failed, what their prompt was, model, tokens, etc. Give me as much information as possible about the failing agent you have access to.
Create 5 sequential 'write' subagents, with each one tasked to create a file with an id that is traceable back to the subagent. Task the subagent with filling the file with lorem ipsum text that is 25kb in size.
On completion of all sequential 'write' agents reporting they have finished creating their files, run 'validation' subagents sequentially to check each file exists and reporting this to you. Each 'validation' subagent should be tracked to match with one 'write' subagent.
On completion of all of the 'validation' subagents, create 'file size' sequential subagents, with the task of reporting the file size to you. Each 'file size' subagent should be associated with one 'write' and their associated 'validation' subagent.
Report the results back to me as the test is performed.
There's a known bug (#4355) regarding parallel agents, but this tests a bug with sequential agents that should avoid that bug.
It can be easily modified to request smaller file sizes (10kb, 15kb, etc) , which seem to lead to more successes. I get consistent failures at 25kb file requests.
If you don't mind, could you throw this prompt at your agent and see if it also fails?
....
Please perform only the following test.
Use sequential subagents only. Do not perform any part of this test yourself and do not call subagents in parallel.
If ANY subagent fails. IMMEDIATELY STOP THE TEST and report the results so far. Inform me which subagent failed, what their prompt was, model, tokens, etc. Give me as much information as possible about the failing agent you have access to.
Create 5 sequential 'write' subagents, with each one tasked to create a file with an id that is traceable back to the subagent. Task the subagent with filling the file with lorem ipsum text that is 25kb in size.
On completion of all sequential 'write' agents reporting they have finished creating their files, run 'validation' subagents sequentially to check each file exists and reporting this to you. Each 'validation' subagent should be tracked to match with one 'write' subagent.
On completion of all of the 'validation' subagents, create 'file size' sequential subagents, with the task of reporting the file size to you. Each 'file size' subagent should be associated with one 'write' and their associated 'validation' subagent.
Report the results back to me as the test is performed.
The AI that actually does things. Emails, calendar, home automation - all from your favorite chat app. New shell, ready to help. The lobster way.
83,815Members
Resources
Similar Threads
Was this page helpful?
