CedarCopilot wants to merge 5 commits into staging from fix/skip-agent-runs-for-warmup-traffic
Live on prod, no production signal yet
OTEL fallback saw 160,224 prod spans for the repo/env since deploy, but 3,483 were errors and the intent has no concrete surface to attribute them to.
Strict CloudWatch fallback found 19 prod failure log lines for [SCHEDULED] Failed to enrich unenriched calendar event (Failed query: insert into \), but this domain-wide failure family is not tied to this intent.
Strict CloudWatch fallback found 1 prod failure log lines for [SUBSCRIPTION] Error setting up Gmail watch (invalid_grant), but this domain-wide failure family is not tied to this intent.
Strict CloudWatch fallback found 6 recent prod failure log lines for [UpdateExternalCrmWorkflow] Driver update failed (fetch failed), but the same failure had 7 log lines in the comparable pre-deploy baseline. Libra is not blaming this PR intent without a post-deploy increase.
Strict CloudWatch fallback found 1 prod failure log lines for [SUBSCRIPTION] Error setting up Gmail watch (invalid_grant), but this domain-wide failure family is not tied to this intent.
Strict CloudWatch fallback found 3 recent prod failure log lines for [backfillSlackChannelHistory] Backfill failed (Failed query: insert into \), but the same failure had 8 log lines in the comparable pre-deploy baseline. Libra is not blaming this PR intent without a.
Libra has verdicts on 0 of 6 tracked behaviors on prod; 6 are still being checked. Libra checks hourly for 3 days after each deploy.
Thirteen fixes across the unibox, the chat artifact surface, the drafting pipeline and the on-event agent. Grouped by what they actually change.
AXIOM_LOGS_DATASET joins SHARED_OBSERVABILITY_SECRET_FIELDS in aws/lib/runtime-contract.ts.
Per that file's own rule, every key there becomes a REQUIRED Secrets Manager field, and ECS
refuses to start any task whose task-def declares a secret missing from the JSON.
Add AXIOM_LOGS_DATASET to the Secrets Manager JSON (an empty string is fine) BEFORE this
deploys, or api/chat/worker all fail to start. Empty is a valid value , the logger falls back
to AXIOM_DATASET, which is today's behaviour.
/inbox/important?channel=all showed 32 archived Rippling receipts. Two independent bugs, one
on each axis of what is really an intersection , "all channels" ∩ "Important".
in:important, which parses to a bare label:IMPORTANT: not
"important mail in my inbox" but everything Gmail ever flagged, archive included. The fallback
now derives from the same base-label map the segment net uses, so the two paths cannot disagree.important/other stubs with
"this tab has no rule at all", letting every chat source dump its unfiltered stream under a tab
whose rule is a Gmail label query they cannot evaluate. They are mail-only by construction now.isLoading: false, so the window before settings resolve read as "loaded" , scopeReady
(built on isPending) holds the query instead of asking it wrong.Every host hides the whole page behind ActiveViewDisplay while an artifact flag is on, so a
branch resolving to null is a blank screen, not a blank panel.
isThreadOpen && !selectedThreadId was exactly how a mention chip blanked it.ArtifactDeadEnd is the floor under all of them.isChannelChatOpen flag, one channelChat
surface, one ChannelThreadView. A LinkedIn chat set as the artifact used to leave every flag
false, so clicking it did nothing at all.inbox.channelItem reads one container's feed row from anywhere, so a chat opened from a
context chip stops rendering its own name over its own id. Authorized by the
channel_container_users join.draft-email emitted <div> 1594/1594
times while chat save-draft emitted <p> on ~40% of drafts, plus a tail of whole
<!DOCTYPE html> documents and self-escaped markup.FontSize/FontFamily on TextStyle, and the Notion
two-column rule no longer claims a 1x2 signature table that carries its own inline widths..draft-body , prose is a dead class
here and Preflight zeroes every <p> margin.next-steps is the only agent that may close a task. As prose the dispatch rule landed at 82%:
173 of 938 outbound sends on deals with an open task skipped it, and 60% of those had a task
marked done by hand later. Biggest driver was [⚙ NOT SIGNIFICANT], read as a blanket bail-out
(45% skip vs 10%) though it exists to collapse the timeline, not to decide whether work happened.
Libra has not measured any production surfaces for this change yet.