mindicator

invAIriant — an evidence-based, multi-lens architecture-audit protocol for LLM coding agents. This skill IS the product: it runs the audit. Commands: audit-pr, audit-range, audit-commit, audit-module, audit-adr, audit-rp, full-audit, verify-findings, classify-severity, synthesize-report, closure-verification. Use when the user asks for an architecture / invariant / lens audit, an evidence-based review, a PR audit, an audit of a commit range, a single commit, a module/directory, ADR↔code drift, or a refactoring proposal, a phase-transition or post-incident audit, to verify candidate findings, classify severity, or synthesize an audit report — or when they type `/invairiant ...`. Every finding must cite concrete evidence: no evidence, no finding.