Files
6eaa5c2c70 fix(cycle): propagate all-provider-failed atom drains so durable jobs retry (#3218)
When every chat() call in a drain batch threw, runBatch mapped only
atoms_extracted/duplicates_skipped (both 0), the loop stopped
'no_progress', and the Minion handler completed the durable job over an
untouched backlog. Derive providerFailure from the phase result
(failures > 0 and zero items processed), surface status
'provider_failure', and throw in the handler so the queue's retry
policy applies; autopilot submits with max_attempts 3 instead of 1.
Partial success unchanged. Takeover of #3248.

Co-authored-by: Masashi-Ono0611 <Masashi-Ono0611@users.noreply.github.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-23 10:57:44 -07:00
..