improve-codebase-architecture
Improve Codebase Architecture
Run an architecture scan grounded in ../coding-standards/SKILL.md. Return one ranked shortlist of refactor candidates within the scan boundary.
Scan, don’t spec. Stop at evidence-backed ownership moves. Do not edit files, run refactors, update documentation, create ADRs, estimate effort, design final interfaces, or write a tech spec. Record migration, compatibility, rollout, or backfill only as current constraints supported by repository evidence or user intent; do not design them during the scan.
1. Set the scan boundary
Use the scope supplied by the user: repository, directory, feature, module, file set, or concern. The boundary controls where candidates may be proposed. Inspect immediate callers, dependencies, composition roots, and tests outside it when needed to establish evidence; record this evidence halo, but do not form candidates for it.
When no useful scope is supplied:
- Inspect the repository shape and major entrypoints.
- Infer the scope only when the repository is small or one area clearly dominates the request.
- Otherwise ask one question that lets the user choose the scope.
Read ../coding-standards/SKILL.md. Search the boundary and its ancestors through the repository root for CONTEXT.md, CONTEXT-MAP.md, equivalent domain-language files, and decision indexes. Search repository documentation and ADR collections for decisions that name or govern the boundary. Inspect every governing source found, plus precedent in the scoped code, tests, and evidence halo.
Completion criterion: The candidate boundary and any evidence halo are explicit; the coding standards are loaded; the prescribed context and decision locations have been searched; every governing source found has been inspected; and relevant precedent has been identified or its absence recorded.