Release 2.1.45 (2026-10-27)¶
Features¶
-
remove the deprecated memory-init, memory-audit and handoff skills (#769) Delete the three skills and their catalogue entries. The recovery-net lines in the issue, epic and operational skills now name the issue's comments as the recovery record and paad:handoff for session continuity. epic-implement points at the worktree-convention spec for the Agent prompt contract, and the three Front-Loaded Judgment references point at the skills reference. The paad policy makes paad:handoff the session-continuity skill and drops its deviation row. The memory-skill list in the CLAUDE.md template region is left for Task 5b. Ref #763
-
add the steering-audit skill (#770) Adds /vergil:steering-audit, the judgment layer of the recurring steering review, checked against the Claude config ownership model page in vergil-tooling. Repo scope (default) inventories CLAUDE.md, .claude/rules/, .claude/settings.json, AGENTS.md and the loadable settings.local.json copies, re-verifies the model's platform facts, runs mechanical checks and a findings-only /paad:pushback, and walks each finding with the human. User scope reviews ~/.claude/CLAUDE.md and ~/.claude/settings.json read-only and writes no file.
Adds the skill to README.md and the skills catalogue, records the findings-only pushback in the paad deviations register, and adds a skill frontmatter check to validate-custom: every skills/*/SKILL.md must have YAML frontmatter with a name equal to its directory and a non-empty description.
Ref vergil-project/vergil-claude-plugin#754
CI Evidence: All gates passed — full audit bundle available. Download →