You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
A file that parsed but declares nothing is no longer a graph gap. Entry
scripts, templates and modules without declarations of their own produced ⚙ ADJUSTED — graph context partial … the index is stale there on every run,
and re-indexing could not clear it. The check now counts the per-file marker
the indexer already writes, so only files the index really lacks are named.
Indexes built before the marker existed keep the old behaviour.
A near-empty agent reply is said on the run. When a model-backed agent
answers a diff of 50+ changed lines with 3 output tokens or fewer on 3000+
input tokens and no findings, the summary carries ⚙ ADJUSTED — <agents> gave a near-empty reply, pointing at a model with
reasoning or the per-agent reasoning setting. The verdict is unchanged. docs/REVIEW_SETTINGS.md explains the knob.
Web
The Repositories page and onboarding say Bitbucket reviews and re-indexes
automatically once the webhook is installed, and manually otherwise.