v3.13.8

v3.13.8 — GraphQL exhaustion resilience + framework-validation fix

This patch release bundles three P0 framework stability fixes that protect the dispatch pipeline under GraphQL rate-limit exhaustion and unblock the entire CI pipeline.

Fixes

• t3025: Guard jq startswith() against null inputs in pulse-merge gates (#21597)
• t3027 (sibling): Idle backoff state machine + telemetry repair + budget-aware prefetch + canonical-recovery (#21596)
• t3027 (wrapper-routing): Route dispatch_with_dedup issue metadata fetch through the gh_issue_view wrapper so REST fallback fires under GraphQL exhaustion (#21606)
• t3032: Repoint MERGE_SCRIPT in test-pulse-merge-approve-collaborator-guard.sh to pulse-merge-gates.sh after the recent split — restored Framework Validation gate (#21609)

Impact

Under GraphQL=0/5000 conditions, the per-candidate dispatch ceremony (claim acquisition, advisory comments, dedup checks) previously failed identically with "unable to load issue metadata" for 100+ candidates per cycle, burning REST/search budget for zero dispatches. After this release:

• Dispatch dedup checks transparently route to REST when GraphQL ≤ 1500 points remaining
• The pulse cycle skips dispatch entirely when GraphQL is below the floor (<5%) instead of attempting and failing
• All workers can resume dispatching as soon as REST budget recovers

Upgrade

aidevops update

Pulse will auto-restart on update. Linux machines: same command.