Disabling tools might be an overkill

[usenkod]

Hi guys,
I had read_file tool failing constantly when I tried to read .yml files from workspace (it was tested to not be a workspace problem). After some testing I found that it can have state "BLOCKED" and it lead to the fact that this tool along with some other tools can be disabled. I set can_use_bash to True for testing at auth.py but that was not the fix.

I found that disabled_tools.update at chat_routes.py had a list of tools to be disabled if agent decided that prompt is an _explicit_web_intent.
I believe that it should not be hardcoded, I would let the agent decide what tools to use for the prompt context. Imho model with good tools understanding would distinguish web intent and other tool calls itself, no need to restrict it.

For the context:

• I wanted to read .yml files (webpage snapshots I saved previously)
• As my prompt contained a web search in one of the steps and those are .yml files my model decided it was web intent even though the instruction was to simply read_file so I could not read those no matter what till I commented read_file at disabled_tools.update
• I am new to this stuff, using native Windows install with gpt-oss:20b at this moment (running it on my 5070 Ti), so I understand it can confuse some instructions and tool usage but still I'd prefer those tools to not be automatically disabled in similar cases