菜单
Designs and implements frontend experiences for agentic applications. Use for chat and non-chat agent UX, streaming state machines, plan and receipt surfaces, approvals, provenance, failure recovery, and trustworthy interaction design.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | frontend-agent-experience |
| description | Designs and implements frontend experiences for agentic applications. Use for chat and non-chat agent UX, streaming state machines, plan and receipt surfaces, approvals, provenance, failure recovery, and trustworthy interaction design. |
| argument-hint | [screen/flow] |
| allowed-tools | Read Write Edit Grep Glob |
Build agentic UI as a high-trust execution console, not a generic text box.
The frontend must make the agent's work inspectable and controllable without overwhelming the user.
The transcript is not the system of record. The system of record is the run timeline and state model.
Do not show only "thinking" or "loading". Show whether the agent is:
Users trust visible evidence:
Make it easy to:
Represent the run as a state machine with at least:
idledraftingplanningexecutingwaiting_for_user_inputwaiting_for_approvalretryingcompletedfailed-recoverablefailed-terminalMaintain separate stores for:
Supports:
Shows:
A chronological event stream with:
Every important step should show:
Must show:
Shows generated files, diffs, previews, or external outputs.
Clearly label:
Never blur "the agent thinks" with "the system confirmed".
For each failure state, show:
At minimum:
When asked for frontend design or implementation guidance, return:
Avoid:
This skill is done only when:
Orchestrates production-grade LLM-driven agent systems in 2026. Use for agent architecture, control loops, topology choice, tool contracts, memory, approvals, safety boundaries, evals, and observability.
Plans and specifies production-grade agent systems. Use when scoping work, writing implementation plans, defining state and tool contracts, sequencing milestones, and producing clear done criteria for LLM-driven systems.
Designs production-grade AI agent experiences. Use for PRDs, user journeys, interaction models, trust and approval UX, failure handling, provenance, and measurable acceptance criteria for agentic products.
Engineers backend systems for LLM-driven agents. Use when designing APIs, orchestrators, tool runtimes, persistence, jobs, auth, retrieval and memory services, approvals, budgets, and reliability controls.
Database engineering skill for agentic systems. Use for schema design, migrations, query correctness, performance, transactions, multi-tenant boundaries, privacy, backups/rollback, and safe DB operations.
Establishes reliability, safety, and release readiness for LLM-driven agents. Use for eval suites, adversarial testing, observability, budgets, policy gates, prompt-injection defenses, and rollout controls.