architecture-review · owainlewis — Claude Code skill: what it does, how to install
architecture-review — a skill for Claude Code and Codex reviews RFCs, ADRs, and technical proposals before code gets written. It flags critical architectural flaws, unsafe patterns, and missing recovery paths.
MIT 347 installs ★ 438 updated 2026-11-10
npx skills add owainlewis/blueprint --skill architecture-review -a claude-codeWhat architecture-review does
The skill evaluates technical decisions early when corrections require minimal effort. The reviewer checks the proposal against existing code, schemas, REQUIREMENTS.md, and ARCHITECTURE.md to spot discrepancies with the live system.
The review traces a concrete request lifecycle from input to output, testing for retries, race conditions, and error states. It evaluates data ownership, schema migrations, rollbacks, and authorization boundaries. Stylistic feedback is omitted entirely: the output only lists blockers and decisions that lead to divergent implementations.
A dedicated read-only agent or an unpolluted subagent (claude -p or codex exec) must execute the review without modifying files. If an independent evaluator is missing, the tool sets the verdict to Blocked.
What it helps with
- catch fatal architecture errors and security bugs before coding begins
- trace request lifecycles through concurrent executions and unexpected failures
- validate database migration plans and rollback feasibility
- simplify technical designs by reducing unnecessary state and moving parts
Example prompts for the agent
- Review the technical proposal in docs/rfcs/002-billing.md for concurrency and data ownership issues
- Run an architecture review on spec.md before we start implementing the changes
How to install architecture-review
1Skills CLI (Vercel)
npx skills add owainlewis/blueprint --skill architecture-review -a claude-code2Manually: put the files in place
~/.claude/skills/architecture-review/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/architecture-review/
1Skills CLI (Vercel)
npx skills add owainlewis/blueprint --skill architecture-review -a codex2Manually: put the files in place
.agents/skills/architecture-review/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add owainlewis/blueprint --skill architecture-review -a cursor2Manually: put the files in place
.cursor/skills/architecture-review/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install owainlewis/blueprint skills/architecture-review2Manually: put the files in place
.github/skills/architecture-review/SKILL.md1Skills CLI (Vercel)
npx skills add owainlewis/blueprint --skill architecture-review -a gemini-cli2Platform CLI
gemini skills install https://github.com/owainlewis/blueprint --path skills/architecture-reviewManually: put the files in place
.devin/skills/architecture-review/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add owainlewis/blueprint --skill architecture-review -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about architecture-review
How to install architecture-review in Claude Code?
Run npx skills add owainlewis/blueprint --skill architecture-review -a claude-code in your terminal.
What does architecture-review do?
It reviews architectural documents like specs, RFCs, and ADRs, returning actionable blockers, important design risks, and a final verdict of Approve, Request changes, or Blocked.
Who actually performs the review?
The review must be conducted by a neutral party, such as a fresh subagent or a separate read-only CLI session (claude -p or codex exec), rather than the agent that drafted the proposal.
Source text
SKILL.md · MIT
--- name: architecture-review description: "Reviews a technical proposal before any code is written: an architecture, spec, RFC, ADR, or design issue. Finds choices that would make it wrong, unsafe, or impossible to prove, and the open questions that block building it." user-invocable: true argument-hint: "[architecture, spec, RFC, ADR, or issue]" disable-model-invocation: true --- # Architecture review Find the choices that would make a design wrong, unsafe, or impossible to prove, while they are still cheap to change. Judge the decisions, not the document's length or format. ## Reviewer Someone who did not write the proposal: a fresh subagent, or another agent CLI run read-only (`codex exec "<prompt>"` or `claude -p "<prompt>"`). If neither is available, say review is blocked. The reviewer changes nothing. ## Steps 1. Read the proposal, `REQUIREMENTS.md`, `ARCHITECTURE.md`, and the code, schemas, and config it touches. Check every claim about the current system against them. 2. Write down the problem, who it affects, the intended outcome, and the main tradeoff. Flag any the proposal leaves unclear. 3. Walk one real request from input to what the user sees, including what happens when a step fails, is retried, or runs twice at once. 4. Check data ownership, interfaces, migrations, rollback, security (auth, untrusted input, secrets), and the scale the proposal claims. 5. Look for a simpler design that gets the same result with less state or fewer moving parts. 6. Check that each important rule and failure has an acceptance check. ## Findings Report only what could change the design: - **Blocker:** unsafe, contradicts the goal or the current system, or cannot recover from a likely failure. - **Important:** two capable engineers could build it in ways that differ for users, data, or security. For each: where it is, what goes wrong and when, the evidence, and the smallest fix or your recommended answer. Say whether you ran it, traced it through the code, or only read the proposal. Skip wording preferences and questions that cannot change the design. No findings is a valid result. ## Verdict End with this block: ```text VERDICT: Approve | Request changes | Blocked UNVERIFIED: <what you could not check, or None> REVIEWED_AT: <commit SHA, or the proposal's date> ``` Any Blocker or Important finding means `Request changes`. Use `Blocked` when context or a reviewer is missing. A detailed document is not a reason to approve. On a later round, confirm the old findings are settled and review only what changed since `REVIEWED_AT`.
© owainlewis · MIT · owainlewis/blueprint
Similar skills
All skills for Code review and refactoring →architecture-review by other authors comes first.