Skip to content

v4.5.1 - Authorization and workflow policy calibration

Choose a tag to compare

View all tags
@oleg494 oleg494 released this 10 Sep 04:34
· 16 commits to master since this release
v4.5.1
e593214

Changes

  • Reconcile core authorization and skill workflows: finish authorized local work without repeated design/plan approval pauses.
  • Remove fixed-count research/retry rituals, mandatory extra workspaces/documents, and missing reviewer-template dependencies.
  • Preserve stop/revocation, read-only scope, and explicit authorization for outward, destructive, spending and memory actions.
  • Keep independent review ownership and scope verification to actual evidence; remove tests that only pinned policy wording.
  • Synchronize all 37 shipped skill mirrors and version metadata to 4.5.1.

Evidence and limitations

  • Local repository suite: 740 passed, 2 skipped, 87 subtests passed.
  • Doctor: all 14 checks green with explicitly expected drift from existing machine-wide skill installations.
  • Dry-run validation: 31 scenarios, 6 tasks and 80 central trigger queries. These validate structure, not model compliance.
  • Earlier calibration: both variants chose the expected action in 12/12 choice cases. In one repeated design-handoff probe, baseline paused for approval in 3/3 trials; candidate continued in 3/3. This is stated-next-action evidence, not an end-to-end coding-quality win rate. Raw calibration evidence remains in the maintainer's local archive and is not shipped.

Installation boundary

This release updates repository sources and shipped mirrors. Existing machine-wide harness installations require a separate deployment. Untracked memory experiments and old build artifacts are excluded.

Archive

coding-kit-4.5.1.tar.gz SHA-256: 7ecf57b87c3ebb2244c0a4d01c6691ffaecd4f87377394dfe643a92622a33922

Assets 3
Loading