Skip to content

kernel_chat v0.6.0

Choose a tag to compare

@GrazianoGuiducci GrazianoGuiducci released this 18 Sep 20:40

kernel_chat 0.6.0 advances the user-owned continuity kernel for cloud chats while preserving a small, present-first host bridge and explicit human authority over external effects.

This release adds and consolidates:

  • capability-relative onboarding: the receiving environment determines which authorized setup effects the assisting model can perform directly and which remain operator-owned;
  • a clearer separation between package source, user-owned instance, configured bridge, operator-confirmed installed bridge, repository reachability, exercised behavior, and later assimilation;
  • optional project/context initialization rather than project identity being required by the kernel;
  • exact bridge byte identity, target and provenance receipts, including conservative handling of legacy or unknown provenance;
  • operator confirmation correlated to the exact bridge incarnation actually delivered, preventing delayed confirmation from being transferred to a newer bridge;
  • temporal receipt validity: historical identity remains distinct from current applicability;
  • cooperative INSTANCE writer conflict visibility without introducing a global transaction manager or controller;
  • fresh remote receipt readback before remote/new-chat reliance;
  • stronger discovery and CommonMark consumer proof using explicit stable anchors and composed anchor→owner relations;
  • preservation of literal user data that resembles template syntax;
  • Python 3.11–3.14 support across Ubuntu and Windows.

Verification for the material rereview:

2c111f2797a7ffe945067ac4b6eee8baace7c9c9

  • external rereview result: no_change
  • CI: 35390696891
  • validator: valid=true, errors=[], warnings=[]
  • tests: 58/58
  • matrix: 8/8

Release-promotion revision:

d2c94ca4e4523f1f7a343601d6beec1e84031fac

  • CI: 35391971484
  • validator clean
  • 58/58 tests
  • 8/8 Python/OS jobs

This release does not automatically migrate existing user instances, replace configured bridges, modify ChatGPT Custom Instructions, prove repository reachability in a specific conversation, or establish behavioral assimilation. Those remain separately selected effects.