Review Architect

What this skill does

• Applies the recognized expert method per subject — recognizable to professionals
• Reviews work and personal matters: texts, code, decisions, processes, data, design, people feedback, reflection, finances/contracts
• Modular: one focused lens per review (10 modules + a generic fallback) — stays sharp, not vague
• Findings with severity (blocking/important/nit) + Conventional Comments label + a concrete fix
• Design: Nielsen heuristics + WCAG 2.2 · Code: Google standard + OWASP/STRIDE · Decision: pre-mortem + red-team
• Process: blameless postmortem + 5 Whys + after-action review · Feedback: SBI + STAR · Reflection: Wheel of Life + GROW
• Clarifies goal, audience & standard up front; better-not-perfect principle
• Safety lenses: fair people feedback without psychoanalysis, no legal/financial advice, kind reflection
• Offers a re-review, an extra lens, a to-do of the blocking points or implementing the fixes
• States review limits (what wasn't checked / assumptions) and gives a clear verdict (approve / approve with nits / revise / reject)

Description

Review Architect is a skill for honest, actionable reviews — code review, design review, UX feedback, decision and process reviews — using the right proven method per subject and delivering severity-ranked findings with concrete fixes instead of sugarcoating. A modular all-purpose reviewer that convinces professionals because every lens builds on the recognized expert method. Shared engine: clarify context & standard, evaluate against explicit criteria, report findings by severity (blocking/important/nit) with a Conventional Comments label, give concrete fixes and a next step — following Google's better-not-perfect principle. Modules with their method: A Text (editing levels, BLUF); B Code & tech (Google code-review standard, Conventional Comments, OWASP Top 10 / STRIDE, ADR & trade-off analysis); C Decision & strategy (pre-mortem after Klein, red-team/devil's advocate, bias check, two-way vs one-way door); D Process & project (blameless postmortem after Google SRE, 5 Whys, after-action review, retro formats under the Prime Directive, actions with an owner); E Data & analysis (bias, correlation≠causation, p-hacking/reproducibility, honest visualization); F Design & marketing (Nielsen's 10 usability heuristics as a heuristic evaluation with severity, WCAG 2.2, AIDA/PAS, one clear CTA, no dark patterns); G People & feedback (the Center for Creative Leadership's SBI model, STAR for a CV); H Personal: reflection & goals (after-action review, Wheel of Life, GROW, SMART); I Personal: finances/contracts/purchases (total cost of ownership/TCO, the fine print — no legal/financial advice); J Generic as a fallback. Honest and constructive throughout (issue, not person), with safety lenses: fair feedback without psychoanalysis, no legal/financial advice, kind self-reflection without harsh self-criticism.

Examples

What it does not do

• No favor-review and no tearing-down — always with an actionable fix
• No legal, financial or medical advice (refers to professionals when stakes are high)
• No diagnoses or psychoanalysis of people, not even in performance reviews
• Won't write/optimize malicious code; recommends no dark patterns
• Won't encourage harsh self-criticism in personal reflection

Compatibility & tech

Tested (internal)

3 scenarios

Recommended runtime

Strongest available model for nuanced, method-faithful assessments.

Modes

Text & document · Code & tech · Decision & strategy · Process & project · Data & analysis · Design & marketing · People & feedback · Personal: reflection & goals · Personal: finances/contracts/purchases · Generic

Inputs

Text/document · Code · Concept/decision · Data/report · Design/screenshot description · Personal reflection

Output format

Structured review: method/lens + standard, overall impression + verdict, strengths, findings with severity + label + fix, 'not checked/assumptions', next step

Subcategory

Review & feedback

License

Proprietär

Security profile

What you get

• review-architect-2.3.0/3 files
  • .claude-plugin/marketplace.json
  • plugins/review-architect/skills/review-architect/3 files

    • SKILL.md
    • manifest.json
    • references/1 file

      • review-architect-checklist.md
  • .agents/skills/review-architect/→ universal — same content (Codex, Cursor, Copilot, Gemini, Windsurf, Cline)
  • LICENSE.txt

Installation

Also works as a chat prompt

Note

Best with a stated goal, audience and desired standard; without these the skill asks briefly instead of guessing. Exactly one method lens is loaded per review.

Changelog

• v2.3.022.06.2026Platform-native output: now actively uses your platform's native formatting — tables, checklists, code blocks and clickable/button options, with a numbered fallback on non-interactive platforms. Plus internal metadata that prepares the upcoming agent builder.
• v2.2.018.06.2026Reworked to the AUTHORING v2 standard: realistic input/answer examples (incl. a calibration case), copy checklist(s) and internal evals; cross-references added.
• v2.1.010.06.2026From test runs (incl. code review): state review limits (what wasn’t checked + assumptions) and clear verdict states (approve / approve with nits / revise / reject; code = LGTM / Request Changes).
• v2.0.010.06.2026Expert methods per module (researched): Nielsen heuristics + WCAG 2.2, Google code-review standard + Conventional Comments + OWASP/STRIDE + ADR, pre-mortem + red-team, blameless postmortem + 5 Whys + after-action review + Prime Directive, SBI + STAR, editing levels + BLUF, Wheel of Life + GROW, TCO. Findings with severity + label, better-not-perfect principle.
• v1.0.010.06.2026Modular review architect: shared engine + 10 module lenses (work & personal) incl. fallback; honest, prioritized findings with fixes; safety lenses.

Frequently asked questions