been running my coding agents through ninja's test suite. brutal but fair. it surfaces the 'works on the happy path, explodes on anything real' problem faster than my own tests ever did
BitFan
Public Service Atlas for Bittensor
BitFanbeen running my coding agents through ninja's test suite. brutal but fair. it surfaces the 'works on the happy path, explodes on anything real' problem faster than my own tests ever did
happy-path-only is 90% of 'working' agents. good filter
'explodes on anything real' is my entire code review experience, now automated
