review-architecture-conformance
Review an implementation, code change, pull request, or deployed design against linked use cases, Business Decision Records, approved ADDs, C4 views, accepted ADRs, quality attributes, and applicable
SKILL.mdskills/review-architecture-conformance/SKILL.md
GitHubReview Architecture Conformance
Workflow
- Read
AGENTS.md, the change diff/code/tests/configuration, and all linked architecture artifacts. - Determine the authoritative baseline: reviewed/approved use cases and BDRs, approved ADD content, accepted ADRs, applicable current/target C4 views, and the project governance profile. Proposed artifacts are advisory only.
- Build a requirement-to-evidence matrix using references/review-method.md.
- Inspect responsibilities, dependencies, interfaces, data flows, trust boundaries, deployment, failure modes, logging, oversight, and test evidence.
- Classify each variance as
Conformant,Documented deviation,Undocumented drift,Artifact stale, orNot verifiable. - Report findings by severity with exact code/artifact references and a concrete remediation. Do not modify code unless separately requested and authorized.
- Create
architecture/reviews/ACR-NNN-<slug>.mdwhen the user requests a persistent review; otherwise return the review in chat.
Rules
- Do not treat an accepted ADR as proof that implementation conforms.
- Do not treat current code as authority that silently overrides an accepted decision.
- Recommend a new/superseding ADR when reality requires a decision change.
- Recommend artifact updates when implementation is correct but documentation is stale.
- Verify applicable privacy, data lifecycle, security, accessibility, safety, transparency, traceability, and human-oversight safeguards.
- Never expose personal/health data, secrets, credentials, or sensitive operational values in findings.
- Never claim approval; identify required owners and decision points.
Output
Lead with overall result and blocking findings. Include baseline artifacts, evidence matrix, findings with severity, compliance assessment, required artifact/code changes, unverifiable items, and recommended owner/action.