roboomp 540eb73bcf fix(advisor): rolled back failed turns before retry or drop ...

Agent.#runLoop appends the user batch and a synthetic stopReason: "error" assistant turn to state.messages before resolving prompt() with state.error set. AdvisorRuntime now snapshots state.messages.length before each prompt, restores it on failure (via a new AdvisorAgent.rollbackTo hook that also resets the advisor's append-only sync cursor), and clears state.error so retries replay a clean baseline and the drop-after-3 path never leaks orphan failed turns into the next successful run's context.

Fixes #3635

2026-06-27 10:46:08 +02:00

..

__tests__

fix(advisor): rolled back failed turns before retry or drop

2026-06-27 10:46:08 +02:00

advise-tool.ts

refactor: renamed search and find tools to grep and glob

2026-06-27 00:57:55 +02:00

emission-guard.ts

fix(advisor): enforced one-advise-per-update + dedupe + noise filter at the enqueueAdvice boundary

2026-06-26 02:48:27 +00:00

index.ts

fix(advisor): enforced one-advise-per-update + dedupe + noise filter at the enqueueAdvice boundary

2026-06-26 02:48:27 +00:00

runtime.ts

fix(advisor): rolled back failed turns before retry or drop

2026-06-27 10:46:08 +02:00

transcript-recorder.ts

feat(coding-agent): supported advisor transcript persistence

2026-06-19 03:50:00 +02:00

watchdog.ts

feat(coding-agent): enabled project context injection for advisor prompts

2026-06-27 06:34:42 +02:00