메뉴
UX design principles for agent-console. Use when designing features, evaluating acceptance criteria, or reviewing user-facing interactions in a multi-agent management UI.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Orchestrator role for strategic decision-making, task prioritization, parallel task coordination via worktree delegation, and first-responder for dev agent questions. Use when managing multiple development agents or making prioritization decisions.
Non-obvious operational details of the local dev environments (single-user `dev.sh`, multi-user `dev-multiuser.sh`, Docker dev stack `docker/docker-compose.yml`) — port layout, source-code propagation, sudo boundaries, multi-repo coexistence. Read when a delegated agent or the Orchestrator first needs to interact with the running dev instance, especially for Browser QA, log inspection, or restart workflows.
Development workflow procedures including conflict assessment, re-implementation proposals, and TDD steps. Use when you need step-by-step workflow guidance beyond what the auto-loaded workflow rules provide.
CodeRabbit code review operations playbook + troubleshooting / FAQ. Use when creating a PR, before merge, when handling CodeRabbit issues (rate-limit fallback, GitHub-side bot unresponsive, both layers simultaneously rate-limited), or when interpreting the 3-layer clean verdict (Pre-merge checks / reviewDecision / inline comments). Covers local CLI invocation, GitHub-side bot interpretation, and case-by-case dispositions.
Detailed test patterns and code examples. Use when you need step-by-step testing guidance, Server Bridge Pattern, or concrete code patterns beyond what the auto-loaded testing rules provide.
Catalog of cross-cutting architectural invariants that code must respect. Use when designing, implementing, or reviewing features that involve shared resources, persistence, or I/O symmetry. Ask each catalog question against the change.
| name | ux-design-standards |
| description | UX design principles for agent-console. Use when designing features, evaluating acceptance criteria, or reviewing user-facing interactions in a multi-agent management UI. |
Five principles for a multi-agent management UI where one user orchestrates many parallel AI agents.
Process multiple items in sequence (A → done → B) instead of forcing return to a central hub between each item. Example: Review Queue (#405) — owner reviews one PR then advances to the next without returning to a queue list.
Surface state changes passively (badges, activity indicators, memos) without requiring active monitoring. Pull, not push — no popups for routine updates. Example: Session sidebar shows agent state (idle/active/asking) via color-coded indicators; agents write memos the user reads asynchronously.
Distinguish urgency: Blockers (agent asking question, error) → prominent indicator, unmissable from any screen. Action needed (PR ready, acceptance done) → badge on next glance. Informational (task completed, CI passed) → passive, on demand. Completions are not blockers.
Actions available where the user already is — inline over page-level, contextual creation, progressive disclosure. Example: Worktree creation from session breadcrumb (#402) — no navigation to a management page.
Summaries in lists, full detail on click. Summarize semantically ("Agent completed PR #42"), not by truncation. Two-level pattern: Level 1 = summary in lists, Level 2 = detail on expand. Example: Session sidebar shows name, state icon, and memo preview without terminal output.
When writing criteria for user-facing features, verify each principle:
See ux-design-standards.md for full rationale, design guidelines, and codebase examples.
