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
{{ message }}
Repository navigation
Commit 9fe2ada
Browse filesBrowse the repository at this point in the historyBrowse files
feat(web): session import keeps going past the 100-conversation batch limit
Each agentSessions.import call imports at most 100 transcripts and 4 GiB per
project. Everything past that budget was counted in skippedCount, so the
welcome wizard reported it as "could not be imported" and users had to reopen
setup by hand, sometimes several times, to finish a large history.
The scanner now reports transcripts left out by the per-attempt transcript or
byte budget as Deferred, and the import result gains an optional deferredCount.
skippedCount stays the total not imported, so older clients behave as before.
Transcripts that were read but could not be parsed are remembered by file
identity, so later attempts skip them without spending the budget again.
The wizard keeps importing a project while each batch leaves fewer deferred
threads, shows per-project progress, and offers Stop importing between
batches. If a batch makes no progress, the rest count as not imported. The
summary separates threads that could not be imported from threads not imported
yet, and offers Continue import when work is left. Servers without
deferredCount get a single batch, as today.
Built with Claude Code (Opus 5.5 orchestrating, Sonnet 5.5 and Haiku 5.5 subagents).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
yield*sql`CREATE TEMP TRIGGER fail_import BEFORE INSERT ON orchestration_events WHEN NEW.event_type = 'message.updated' BEGIN SELECT RAISE(ABORT, 'test publication failure'); END`;
0 commit comments