Creator

Architecture Review (review-architecture) is a reusable skill for reviewing software designs, schema mappings, service boundaries, module structure, and generated code. It captures my approach as the architect overseeing implementation.

A skill is a folder of instructions and supporting references that ChatGPT or Codex reads when performing a task. This one is instruction-based; it does not run a separate review server.

What it reviews

  • Preserve authoritative models, such as CWM, rather than inventing ad-hoc schema.
  • Keep transport, domain rules, persistence, orchestration, and presentation responsibilities clear.
  • Favor clean distributed patterns where they provide useful ownership and reuse.
  • Inspect generated code against the intended architecture, with concrete evidence and bounded recommendations.
  • Review ZMQ contracts, socket ownership, timeouts, retries, lifecycle, and pluggable service interfaces.

CWM applies where the project adopts it. The skill does not impose CWM on every application. Reviews remain read-only unless implementation is also requested.

Where it lives

Environment Location Meaning
ChatGPT Architecture Review in my skill directory Installed personal skill.
ChatGPT hosted execution environment /root/.codex/skills/remote-skills/skill-6ac12d7cba4881919f71c849532a07b5/ Hosted source folder; this is not a path on Sally or Wintermute.
Local Codex / VS Code ~/.agents/skills/review-architecture/ User-wide destination when installing the exported skill locally.
One repository only <repo>/.agents/skills/review-architecture/ Alternative repository-scoped destination.

The local destination is the installation path to use; this page does not confirm that the ZIP has been extracted on a particular machine. In VS Code Remote SSH or a container, install it in the environment where Codex executes.

The folder contains:

File Purpose
SKILL.md Review instructions and skill name/description.
references/zmq-patterns.md Repository-backed ZMQ patterns and review guidance.
agents/openai.yaml Skill presentation and invocation metadata.
assets/icon.svg Skill icon.

Use it in ChatGPT

Open the installed skill from the directory link above, or type @ in the composer and select Architecture Review.

Example request:

Use Architecture Review to review this proposed design.
Check responsibility boundaries and fidelity to the authoritative model.
Give evidence-backed findings and the smallest useful corrections.
Keep this review read-only.

Provide the design, repository, relevant files, or diagram to inspect. Identify the governing specification and version when model conformance matters.

Install it for local Codex / VS Code

Download the previously exported review-architecture.zip from the ChatGPT conversation. The export includes the ZMQ reference.

For a user-wide installation, assuming the ZIP is in Downloads:

mkdir -p ~/.agents/skills
unzip ~/Downloads/review-architecture.zip -d ~/.agents/skills
test -f ~/.agents/skills/review-architecture/SKILL.md
test -f ~/.agents/skills/review-architecture/references/zmq-patterns.md

Check the extracted layout: SKILL.md must sit directly inside review-architecture/, not inside an extra nested folder. Install the complete folder so relative references remain available.

For a repository-scoped installation instead, run from that repository’s root:

mkdir -p .agents/skills
unzip ~/Downloads/review-architecture.zip -d .agents/skills

Choose the user-wide or repository location to avoid duplicate entries with the same skill name.

Use it in the Codex extension or CLI

Open the project in the Codex extension. Type $ to select review-architecture, or use /skills to find it.

Code-structure review

Use $review-architecture to review this repository's code structure.
Trace a normal operation and a relevant failure path.
Identify mixed responsibilities, hidden coupling, and duplicated policy.
Keep this review read-only and cite the relevant code locations.

ZMQ service review

Use $review-architecture to review this ZMQ service.
Check whether a new service can register handlers and supply configuration
without editing transport code. Review message contracts, correlation,
timeouts, retry ownership, socket lifecycle, and control/event separation.

CWM mapping review

Use $review-architecture to compare this schema with OMG CWM 1.1.
Map implementation entities to metamodel elements and verify inheritance,
associations, derived properties, and cardinalities from the specification.

Explicit invocation makes the intended review workflow clear. The skill can also be selected automatically when a task matches its description.

Expected output

A review should explain what works, prioritize material findings, identify the evidence and practical consequences, and recommend the smallest useful changes. It should distinguish architectural defects from stylistic preferences and state what was verified or remains uncertain.

The ZMQ guidance combines R3el’s injected routing, Ax3l/R3el’s bounded requests without automatic retries, and Snake Lab’s correlated contracts and separate channels. Its pinned examples are historical evidence; current project code should be inspected before applying them.

Updates and troubleshooting

  • If the skill is missing locally, verify SKILL.md exists at the expected path and that Codex is running in that same environment.
  • Check the skill selector for duplicate installations or disabled skills.
  • A local ZIP installation is a separate copy. Re-export and replace that copy when the ChatGPT skill changes; do not assume automatic synchronization.
  • If a reference cannot be loaded, check that references/zmq-patterns.md was installed alongside SKILL.md.

Installation and invocation guidance checked on October 3, 2026.