mcpField
A
architectural-analysis

architectural-analysis

github.com/pedronauckTesting & quality644533 installsUpdated 1w ago
Repository

About architectural-analysis

architectural-analysis is an agent skill for auditing software design within a requested scope. It prioritizes evidence-backed findings, accounts for project boundaries and compatibility, and can guide remediation when authorized.

Key features

  1. 01Scope audits to relevant entry points and dependencies
  2. 02Classify design issues using a detection catalog
  3. 03Trace usage before flagging code as dead
  4. 04Cite evidence, uncertainty, and minimal changes
  5. 05Adapt report depth to the request

Use cases

  1. 01Review architecture before a planned refactor
  2. 02Assess dependencies and public exports in a scoped audit
  3. 03Document evidence-backed design findings
  4. 04Apply authorized remediation for confirmed issues

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/pedronauck/skills --skill architectural-analysis

After installing, type/architectural-analysisin Claude Code to confirm it loaded.

Bundled files

References and scripts installed along with SKILL.md.

assets/
report-template.md
summary-template.md
references/
detection-catalog.md

Details

From the project README

Architectural Analysis

Audit the requested scope and report concrete design problems. An audit alone is read-only; existing explicit authorization to apply findings can include remediation. Use the project's boundaries and public compatibility policy when judging a design.

  1. Resolve scope from the request/context. Map entry points and relevant dependencies with rg --files, scoped rg searches, and available static-analysis output. Do not create one todo per source file or assume a TypeScript stack.
  2. Prioritize signals: dead exports, duplicated responsibilities, cycles, confused ownership/types, and expensive change patterns. Read the matching section of references/detection-catalog.md when classifying a candidate; thresholds are clues, not automatic defects.
  3. Trace usage before declaring code dead, including dynamic loading, reflection, framework hooks, tests, and published APIs. An unreferenced public export is not safe to delete solely because this checkout has no consumer.
  4. Confirm each material finding in source. Explain the behavior or maintenance cost, cite paths/lines, state uncertainty, and propose the smallest useful change. Avoid a separate generic smell sweep when it adds no evidence.
  5. Match report size to the request. For a broad audit, use assets/report-template.md as an adaptable outline and save .audits/architectural-analysis-<timestamp>.md; distinguish inspected and uninspected areas. For a narrow answer, a concise finding list is enough. Missing coverage must never be reported as None found.
  6. Summarize the important findings and evidence. Existing source and summary templates are optional presentation aids, not two required reports.

Comments

Markdown supported