What problem does it solve? Architecture Decision Records often become zombie documents that nobody enforces. This Skill writes ADRs that distinguish mechanical constraints (verified automatically by a verifier running grep/lint rules) from discursive ones (verified by code review), so decisions stay binding throughout the build. ## Core Features & Use Cases - Executable enforcement rules: Attaches an enforced_by grep/lint rule to mechanical constraints, with POSIX-portable, shell-safe, comment-stripped patterns validated against positive and negative fixtures. - Prohibition vs presence rules: Supports wave-gated presence rules (exigible_from: <block-id>) so a rule only becomes enforceable once the owning block is merged, avoiding false reds. - Consistency reconciliation: Closes answered spikes in the context-map, supersedes replaced ADRs, and checks decisions against profile boundary rules before finalizing. - Use Case: During the model movement of a feature, record the decision to forbid a specific HTTP client library; the Skill emits decisions/0007-no-retrofit.md with an import-anchored grep rule the verifier runs on every diff. ## Quick Start Ask the agent to write an ADR for a specific architectural decision, stating the scope, whether the constraint is mechanically enforceable, and any spike it closes.