Video unavailable. Open the source below.
NewLLM Debugger
A VS Code debugging extension uses Jev to route each debugging step, with a language model available as a baseline or for other tasks.
Added to Jevfast
How it uses Jev
The choice routes a debugger step. `readyForFix` at probability 0.7 or `finish` choice confidence 0.75 ends the hunt. Parameterless steps execute directly; Jev-selected breakpoint/evaluate actions invoke the generation model to supply arguments.
What Jev decides
Illustrative source-defined Jev questionExample answers · not a recorded Jev response · Source ↗
Question 1 · nextAction
YOUR APP
INSTRUCTION
Which single debugger action best advances the hunt for the failing check's cause?
STATE
State is `renderObservation(observation)`, the formatted current debugger observation including runtime state.
JEV · CHOICE
- next
- stepIn
- stepOut
- continue
- setBreakpoint
- evaluate
- finish
App workflow
- Observe the paused program
Read the debugger observation, including runtime values and execution context.
- Choose and validate an action
Jev selects the next debugger action and evaluates readiness for a fix. Code validates the action and uses a helper model when action arguments are needed.
- Execute and repeat
Execute the debugger action, observe the result and continue the bounded debugging hunt until its completion conditions.
Why it is interesting
It treats debugging as a sequence of constrained routing decisions and makes a language-model routing baseline available for comparison.