菜单
Define or revise a documented reusable engineering mode using native Claude Code Skills and custom agents. It does not create JavaScript workflow runtimes.
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
Screenshot-first visual end-to-end validation policy for browser-rendered UI and webview jobs after verifier approval.
Default engineering workflow. Classify a technical request, route it through the native T1/T2/T3 engineering fleet, and return only independently verified completion to the orchestrator.
Safely operate Claude Code's opt-in advanced capabilities—sessions, worktrees, agent teams, Skills, MCP, plugins, hooks, Channels, schedules, goals, and Agent SDK integrations—when the user explicitly asks for one. Do not use for an ordinary build request.
Routes selective local Codex plan, code, verification, and visual judgment with evidence-backed blocking and strict cost caps.
Shared evidence, accessibility, context, and handoff rules for the local design-agent fleet.
Produces contextual creative theses and bounded visual directions without generic model defaults.
| name | workflow-author |
| description | Define or revise a documented reusable engineering mode using native Claude Code Skills and custom agents. It does not create JavaScript workflow runtimes. |
| disable-model-invocation | true |
| argument-hint | <workflow purpose> |
Use this only on an explicit request. Before writing a mode, establish its name, trigger, inputs, roles/models, write ownership, allowed parallel branches, outputs, success condition, maximum retries, failure behavior, and whether it changes files. A reusable route cannot depend on an unanswered material mid-run human question.
Add or revise a Skill and only the custom agent definitions required for the mode. Do not create .claude/workflows/*.js or invent unsupported phase, log, agent, or pipeline APIs. Keep the skill as instructions and structured handoff contracts that the interactive orchestrator executes.
Keep the mode narrow and reproducible. Use structured handoffs, bounded retries, and explicit terminal status (completed, blocked, or failed). Do not replace an ordinary interactive /build task with a hidden runtime. Do not embed claude -p calls; if programmatic orchestration becomes essential, use the Claude Agent SDK's interactive streaming/session APIs and obtain explicit user approval first.
After implementation, add static routing fixtures and document the trigger, inputs, side effects, and recovery steps in docs/CUSTOM-WORKFLOWS.md.
