name: audit-architecture description: Audit Architecture, Scalability & Availability. Use when auditing ADRs, interface contracts, coupling, scalability, resilience patterns, SLA/RPO/RTO, high availability, or incident management. Triggers on "audit architecture", "audit archi", "audit scalabilité", "audit scalability", "audit disponibilité", "audit availability", "audit résilience", "audit resilience", "audit incidents". version: 1.0.0 metadata: last-reviewed: 2026-06-19 category: audit

Audit Architecture, Scalability & Availability

The project variant (<repo>/.claude/skills/) wins if it exists: it tools this generic protocol with scripts, datasets and CI gates specific to the repo.

When ME and not audit-industrialisation

• ME when: audit targeted at this domain only
• audit-industrialisation instead when: a global multi-domain audit - it orchestrates every audit (ME included) and produces a consolidated report

For a full project audit, use audit-industrialisation rather than invoking each skill separately.

Overview

Audit a project's architectural maturity: documented decisions, API stability, coupling, resilience patterns, scalability, availability, operational support and incident management. Produces a maturity score (0-4) per question.

74 questions split across 7 sections:

The detailed grids (statements, criticality, conditions, verification commands, levels 0-4) live in reference/ - one file per section.

Score calculation

• Each scored question is assigned a level from 0 to 4.
• Questions marked N/A are excluded from the calculation.
• Global score = sum of levels / number of scored questions (excluding N/A).
• Formula: score = Σ(levels) / count(scored questions)
• Example: 50 questions scored out of 74, sum = 120 → score = 120/50 = 2.4/4

Audit protocol

1. Detect the project type: backend/API, frontend SPA/PWA, fullstack, library/SDK. Adjust the N/A questions per type (see Variants below).
2. For each section (AR, SC, PA, RE, DI, SU, IM in this order):

• Read the matching reference/<section>.md file.
• For each question: run the listed verification commands.
• Assign a level 0-4 (or N/A) in light of the grids, with justification and confidence level.
• Handle the conditional questions (e.g. AR-01a is only evaluated if AR-01 >= 2): if the condition is not met, mark N/A.

1. Mark N/A the questions that do not apply to the project type.
2. Compute the global score excluding N/A.
3. Produce the report via audit-report.

Variants by project type

Frontend SPA/PWA - adapt:

• SC-01 to SC-07: PWA offline, service worker, bundle size, CDN, lazy loading. No HPA/K8s.
• PA-01: error boundaries (React/Vue), offline fallback, service worker cache, retry on API calls.
• PA-06: service worker lifecycle, graceful update.
• PA-12: offline fallback, stale-while-revalidate, IndexedDB cache.
• PA-18 to PA-21: usually N/A (no frontend DB).
• RE-01: retry safety of client-side API calls.
• DI-01 to DI-09: CDN availability, service worker health, connectivity detection.
• IM-01 to IM-09: error tracking (Sentry), performance monitoring (Web Vitals).

Library / SDK - usually N/A: SC-01 to SC-07, PA-05 to PA-09, PA-12, PA-18 to PA-21, RE-05 to RE-08a, DI-01 to DI-09, SU-01 to SU-03, IM-01 to IM-09. Focus on AR-01 (ADR), AR-02 (tech debt), AR-07 (backward compatibility), AR-08 (deprecation), AR-09 (semver), PA-01/PA-03 (retry guidance in the docs).

Backend / API - every question applies with the standard grid.

Grids by section

Exit condition

• [ ] All 7 sections have been covered (each reference/<section>.md read).
• [ ] Every question is scored 0-4 or marked N/A (conditional questions included).
• [ ] Every level rests on evidence from running the verification commands.
• [ ] Global score computed excluding N/A per the formula.
• [ ] Report produced via audit-report (summary, per-question detail, strengths/weaknesses, MUST/SHOULD/COULD recommendations, non-auditable items).

Changelog

• 1.0.0 (2026-06-19) - initial versioned release + state-of-the-art enrichment (routing, context, protocol, traps, exit condition); audit grids in reference/ (progressive disclosure)