diogox451/principal-software-architect — explained in plain English
Analysis updated 2026-05-18
Get an AI-generated architecture decision record for a hard-to-reverse system design choice.
Have the AI run a severity-ranked review of your codebase's architecture against established patterns.
Ask whether to split a service, backed by references to Clean Architecture, design patterns, and microservice best practices.
Run a multi-agent validation where one agent proposes, another critiques, and a third synthesizes a final verdict.
| diogox451/principal-software-architect | hwdsl2/docker-litellm | dcarrero/awesome-code-review-prompts | |
|---|---|---|---|
| Stars | 9 | 9 | 8 |
| Language | Shell | Shell | Shell |
| Last pushed | — | 2026-07-09 | — |
| Maintenance | — | Active | — |
| Setup difficulty | easy | moderate | easy |
| Complexity | 3/5 | 3/5 | — |
| Audience | developer | pm founder | developer |
Figures from each repo's GitHub metadata at analysis time.
Install with a single skills.sh command or clone and run the bundled install script.
Principal Software Architect is an installable skill for AI coding assistants that makes the assistant behave like a senior software architect during conversations. It works with Claude Code, Codex, Cursor, OpenCode, Grok, and other agents that support the open Agent Skills format, and it can be installed through the skills.sh command line tool or by cloning the repository and running an install script. Once active, the skill asks the AI to think through constraints and existing patterns before suggesting a new structure, to compare design options by weighing the cost of reversing a decision, and to prefer extending the lowest existing layer of a codebase rather than introducing a brand new parallel framework. It can write architecture decision records for choices that are hard to undo, and it produces structured review output that flags issues by severity. The skill grounds its judgments in written summaries of several well known software books: Robert Martin's Clean Architecture, the Gang of Four's Design Patterns, Martin Fowler's writing on production ready microservices, Clean Code, and a book on algorithms called Entendendo Algoritmos. Each of these is distilled into short rule codes, for example CC for Clean Code rules or ALG for algorithm rules, so the AI can point to a specific standard when it flags a problem instead of giving vague advice. For each piece of code or architecture it looks at, the skill outputs a short structured line summarizing what was expected, what was actually observed, whether it fits established patterns, and what to do next. It can also run a multi agent check, where one agent proposes an assessment, a second agent argues against it, and a third agent produces a final synthesis before giving a verdict. Additional specialist checklists cover domain modeling, distributed data, API contracts, reliability, security, and how architecture decisions are documented and communicated to a team. The project is released under the Apache 2.0 license and includes both English and Brazilian Portuguese documentation. The package is meant to be installed once and then invoked either with a slash command or by describing the kind of review or decision you need in plain language.
An installable AI agent skill that turns Claude Code, Codex, Cursor, or Grok into a senior software architect for reviews, ADRs, and design trade-offs.
Mainly Shell. The stack also includes Shell, Claude Code, Codex.
Apache 2.0 license, use freely including commercially, with patent grant and attribution requirements.
Setup difficulty is rated easy, with roughly 5min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Don't trust strangers blindly. Verify against the repo.