Commit 50ca97d
committed
fix(steering): don't inject interruptions after final response
Remove the drain-after-final-response logic that forced the loop to
continue after the LLM had already finished. This caused confusing
double responses.
Correct model:
- Tool execution phase: interruption = steering (single merged response)
- Final response phase: new message = new task (separate responses)
Orphaned messages left in the checker queue after the task completes
are now re-published to the inbound bus so they're processed as
independent new tasks by the run() loop.
Made-with: Cursor1 parent ec5022a commit 50ca97d
1 file changed
Lines changed: 6 additions & 11 deletions
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
417 | 417 | | |
418 | 418 | | |
419 | 419 | | |
420 | | - | |
421 | | - | |
422 | | - | |
423 | | - | |
424 | | - | |
425 | | - | |
426 | | - | |
427 | | - | |
428 | | - | |
429 | | - | |
430 | 420 | | |
431 | 421 | | |
432 | 422 | | |
| |||
559 | 549 | | |
560 | 550 | | |
561 | 551 | | |
562 | | - | |
| 552 | + | |
| 553 | + | |
| 554 | + | |
| 555 | + | |
| 556 | + | |
| 557 | + | |
563 | 558 | | |
564 | 559 | | |
565 | 560 | | |
| |||
0 commit comments