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
Shipped in a stronger form by the knowledge-intake loop (RFC record: #111, doc/RFC-knowledge-intake.md; code: zuiho-kai/omni-reviewbot#21/#22 + #110, all merged and deployed 2026-08-28):
"weekly AI update" → a daily distillation batch: every merged vllm-omni PR and every copilot bugfix run is recorded, distilled into owner-scoped executable rules (validators fail-closed), and delivered as one PR against this repo from the bot's fork.
"monthly owner confirm" → confirmation moved to per batch: the owner reviews and merges each knowledge PR, so nothing enters the tree unconfirmed. The monthly cadence survives as the quality probe (eval: monthly review-quality probe of omni-reviewbot as a harness arm #109's run_reviewbot_monthly.sh) that re-judges the benchmark PRs against the pinned baseline.
Enablement is operational (KNOWLEDGE_INTAKE_ENABLED → fork + KNOWLEDGE_PR_ENABLED), per the RFC's rollout section.
Activity
tzhouam commentedon Aug 28, 2026
Shipped in a stronger form by the knowledge-intake loop (RFC record: #111, doc/RFC-knowledge-intake.md; code: zuiho-kai/omni-reviewbot#21/#22 + #110, all merged and deployed 2026-08-28):
run_reviewbot_monthly.sh) that re-judges the benchmark PRs against the pinned baseline.Enablement is operational (KNOWLEDGE_INTAKE_ENABLED → fork + KNOWLEDGE_PR_ENABLED), per the RFC's rollout section.