Add lazarus-forge (design-review gate) and audit-repair skill

[CognitiveCodeAI]

What this adds

1. lazarus-forge — a new sibling plugin with the design-review skill: a bounded, evidence-backed, runtime-verified design-review quality gate for Claude Code skill/plugin/agent/MCP/hook proposals. Emits exactly one verdict (BUILD / BUILD-WITH-CHANGES / DO-NOT-BUILD / NEEDS-MORE-DETAIL) with a closed required-changes list so reviews converge instead of looping. Ships the rubric, report template, standards registry, and a labeled gold-set fixtures file.

2. audit-repair — the strategic apply skill in core lazarus (the audit → audit-repair counterpart to discover → repair): executes a ratified CODEBASE_AUDIT.md §11 one finding at a time — ratify → act → verify against each item's acceptance check — with safety-rails-first ordering, namespaced AUDIT_* forensic artifacts (no collision with repair's), honest Bash-only-guard scoping, and one-way read-only issues interop.

Plus: a single canonical slug algorithm (fully-lowercase slugify) pinned in lazarus-github:issues and referenced by audit-repair, with worked examples; and the gated design-of-record for audit-repair in docs/.

Verification done

• claude plugin validate passes on both plugins (only the expected no-version warning).
• lazarus-forge:design-review self-dogfooded (F0 → BUILD) and gold-set calibrated (F0–F7 green).
• audit-repair was reviewed through design-review (verdict BUILD; adversarial pass caught and folded in 4 corrections: filename collision, guard-scope honesty, assertion-PASS evidence rule, §9/§10 DoD).
• All five arXiv citations independently verified real.

Merge note

Merging publishes both to the cognitivecode marketplace (served from main).