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
This commit was created on GitHub.com and signed with GitHub’s verified signature.
[0.1.38] — 2026-07-14
Upgrade notes
The local daemon protocol is now v7. Upgrading may briefly leave one v6 and
one v7 daemon, plus their warm LTM children, running for the same logical
configuration. With the default 900-second idle timeout the v6 daemon exits
naturally; pinned daemons (idle_timeout_seconds=0) remain until mms daemon stop --all. (#694)
Added
Context-compose schema 3 preserves budgeted adjacent chunks from compatible
cores across direct and shared-daemon routes. Schema 2 remains supported for
scoped Pinned Context without a visible context-window guarantee. (#694)
Schema 3 window requests receive a larger wire-only retrieval budget while
the configured model-injection limit remains unchanged. (#694)
Added a copy/paste reviewed-project-resume guide covering project-local
Pinned Context, automatic surfacing, nearest-neighbor previews, and the
separate review-first proposal boundary. (#694)
Changed
Daemon protocol v7 negotiates a client maximum and selected compose schema
per response. The protocol bump deliberately prevents a stale v6 daemon from
silently suppressing schema 3 until restart. (#694)
Schema 3 decoding now retains at most ten adjacent chunks per direction,
preserving the nearest chunks and ignoring malformed overflow that STM will
never render. (#695)
Core compatibility advisory now pins 0.3.8 legacy, 0.3.9 schema 2, and 0.3.10
schema 3 as separate scheduled cells. Its schema 3 cell also runs the
reviewed-project-resume core CLI and candidate-review contract in a temporary
Git project. (#695)