# X9 Architecture Scout

> Use when the user asks for a read-only architecture audit that compares plausible directions and makes component boundaries, dependencies, ownership, coupling, or change hotspots visible — «проведи архитектурный аудит», «покажи проблемные места архитектуры», "audit this codebase architecture". Do not use to implement a refactor, create a general implementation plan, or merely explain one file.

- Skill: `xonika9/x9-architecture-scout` (Agent Skill, multi-file: 5 files)
- Install (CLI): `npx skillmds@latest add xonika9/x9-architecture-scout`
- Raw SKILL.md: https://api.skillmd.com/api/skills/xonika9/x9-architecture-scout/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Web & Frontend
- Author: xonika9 (https://skillmd.com/u/xonika9)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/xonika9/x9-architecture-scout

---


# Architecture scout

Produce a read-only architecture report for a repository or user-named subsystem. This skill is model-invoked in Claude Code and Codex; its portable claim is structural Agent Skills compatibility only. Evidence tier: static.

Inspect component boundaries, dependency direction, ownership of data and behavior, coupling, duplication, change hotspots, and seams. Treat source, configuration, tests, and version history as evidence; distinguish observed facts from inferences. Do not edit production code, refactor, install dependencies, copy an external skill, or change external systems.

## Method

Discover `codebase-design` from the live current-runtime catalog before any project or user/plugin roots that runtime declares. Resolve candidate paths symlink-aware, read the live skill version when one is found, and do not retain, reproduce, or hard-code its machine path. Use its method only to enrich the audit.

If no readable live `codebase-design` is found, perform the direct baseline audit and mark the method status `DEGRADED`. If the user requires Matt's exact method, stop with method status `BLOCKED`; do not substitute an invented version. Read [the report contract](references/report-contract.md) before creating an HTML report; read `x9-diagrams`' Mermaid and design references when a diagram is needed.

Follow the repository's report-location convention. When it has none, write one HTML file at `docs/architecture-reviews/YYYY-MM-DD-<stable-slug>.html`. A same-day fallback audit must not silently replace a prior report: re-read and replace a named existing report only for an explicit update; otherwise choose a non-colliding stable suffix or path, or ask one short question when the report identity matters. The report may recommend a candidate, but it does not implement it; after a selection, hand the decision to `ce-plan` when available or provide a portable planning handoff.

## Evidence and handoff

Run `python3 scripts/validate_report.py <report.html>` from this skill directory against the actual report. It proves structural and accessibility-contract compliance only. Keep the validator result separate from the actual render status: only a browser check at `1280px` and `390px` can make rendering `PASS`; otherwise state `DEGRADED` or `NOT_PROVEN` with the missing evidence.

Completion is one report that validates and makes the decision question, method/structure/render statuses, conclusion with confidence, candidate comparison, evidence, diagrams, risks, and next step inspectable without JavaScript. Report the artifact path, all three statuses, source scope, principal evidence, and the planning handoff or blocker.

The [onboarding declaration](references/onboarding.json) lists external prerequisites.