We want to control the safety level of the tools when used by LLMs in agentic tooling loops. Potential options:
--safe flag that disables all destructive actions--safety-level flag that allows for levels (NODELETE, NOUPDATE, etc)- env var support
- http middleware to block actions
flag should be configurable at a global level (env var, config, etc) and not easily ejected by the LLM (perhaps standard CLI flags should not be used for this reason).
We want to control the safety level of the tools when used by LLMs in agentic tooling loops. Potential options:
--safeflag that disables all destructive actions--safety-levelflag that allows for levels (NODELETE, NOUPDATE, etc)flag should be configurable at a global level (env var, config, etc) and not easily ejected by the LLM (perhaps standard CLI flags should not be used for this reason).