What problem does it solve?
Designing a new feature or architecture for the Crowi wiki requires deep codebase research, prior-art analysis, adversarial review, and careful documentation — work that overloads a single AI session and often produces shallow specs. This Skill orchestrates that entire pipeline while keeping the main conversation lightweight.
Core Features & Use Cases
- Two-phase Workflow orchestration: Phase A researches the codebase, prior decisions, and external prior art in parallel, then synthesizes 2-3 design approaches with an RFC-vs-spec recommendation; Phase B writes the document and runs adversarial multi-lens reviews with a bounded revise loop.
- Human decision gate: Design convergence (approach selection, RFC vs spec, open questions) happens through an interactive gate, so judgment stays with the user while heavy lifting runs in subagents.
- Codex offload with fail-open fallback: Analysis, writing, and review stages run through codex-run.sh on tiered models and automatically fall back to Claude agents when Codex is unavailable, with every fallback reported.
- Use Case: Ask to design a new Markdown notation for Crowi; the Skill researches the renderer pipeline, proposes approaches, lets you pick one, then produces an adversarially reviewed implementation-ready spec that the crowi-feature skill can implement directly.
Quick Start
Ask the AI to run /crowi-design followed by your design topic, for example to design image display attributes for the wiki renderer.