openclaw - 💡(How to fix) Fix Severe regression from 2026.3.13 to 2026.4.15: web actions degraded, image grabbing worse, tasks often go silent [1 participants]

Summary

I upgraded from 2026.3.13 to 2026.4.15, and the regression is very obvious.

After the update, the assistant became dramatically worse. This is not a subtle change, it feels like a major capability regression and is now barely usable for normal work.

Regressions observed

Compared with 2026.3.13, several tasks that used to work with simple instructions now frequently fail or never complete:

• Opening and navigating web pages is much worse than before
• Extracting or grabbing images from web pages has regressed significantly
• Tasks that used to be handled directly now often fail to execute

Most frustrating behavior

The worst part is not only failure, but the lack of timely feedback.

A common pattern now is:

• I give a task
• the assistant appears to start
• then nothing happens
• no useful progress update
• no clear error
• no completion
• it just goes silent

This "black hole" behavior happens repeatedly and makes the product very hard to use in practice.

Expected

A version upgrade should maintain or improve capability and responsiveness.

At minimum, if a task cannot be completed, the assistant should provide timely feedback, clear failure signals, or actionable error reporting.

Actual

2026.4.15 feels like a clear downgrade in both execution ability and interaction quality:

• weaker web interaction capability
• worse image grabbing from pages
• more failed simple tasks
• frequent silent stalls with no timely feedback

Please investigate

Please check whether 2026.4.15 introduced regressions in:

• browser or web automation behavior
• image extraction / page parsing flows
• tool-calling reliability
• model routing or default model selection
• runtime/task status reporting
• timeout / failure handling
• progress feedback behavior

Additional note

I can provide before/after examples, exact logs, and specific failed tasks if needed.