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。
直接命令不会经过审查 Prompt;运行前请先检查来源。
npx skills add https://github.com/Marvino-Fransisco/agent-registry --skill create-agent命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地?可下载 SkillsMP 当前能够提供的文件。
Use this skill when working on tasks, answering questions, or brainstorming related to backend systems and engineering. Focus on architecture, project setup, system design patterns, and code quality standards.
Create and configure new custom AI skills with specific standards, references, and instructions. Use this for requests like "make a skill", "create a custom skill", "build a new skill", or "set up a new skill".
Use this skill when working on tasks, answering questions, or brainstorming related to data structure, schema design, and data flow.
基于 SOC 职业分类
正在显示 SKILL.md
| name | create-agent |
| description | Use this skill when you are creating new agent and editing an existing agent. |
To create new agent, use the following rules, variables, and procedure:
These rules are non-negotiable.
When in doubt, stop and ask. Proceeding with uncertainty is always worse than pausing.
You must stop and escalate to the user when:
You must never:
Every output must be traceable, consistent, and stored in the correct location.
{agent-name}.md
Example: data-designer.mdThe variable will be called via {{VARIABLE_NAME}}
These are global variables.
| Variable | Value |
|---|---|
| CLAUDE_GLOBAL | ~/.claude/agents |
| CLAUDE_LOCAL | ./.claude/agents |
| OPENCODE_GLOBAL | ~/.config/opencode/agents |
| OPENCODE_LOCAL | ./.opencode/agents |
| TEMPLATE | template |
| EXAMPLE | example |
| CLAUDE_YAML_TEMPLATE | claude_template |
| OPENCODE_YAML_TEMPLATE | opencode_template |
| REGISTRY | ~/agent-registry/agents/{{USER_PLATFORM}} |
These are global variables, but you must fill it by yourself
| Variable | Value |
|---|---|
| USER_PLATFORM | empty |
| AGENT_SCOPE | empty |
Read all the knowledge needed
Required output at the start of this phase:
Working on Phase 1 - Knowledge fill
Required output before proceeding:
"Assets read [list]"
Proceed to Phase 2
Ask as detail to user
Required output at the start of this phase:
Working on Phase 2 - Context gathering
Constraint:
Required output before proceeding:
"Agent detail [list]"
Proceed to Phase 3
You must reflect on Phase 1 and 2 by answering every verification question honestly. Saying "yes" without evidence is itself a violation.
Required output at the start of this phase:
Working on Phase 3 - Reflection
Answer every question below explicitly. Do not skip any. Do not answer with a generic "yes". Provide evidence for each answer.
Phase 1 Verification:
assets files, or did I guess them from memory?Phase 2 Verification:
Required output before proceeding:
"Reflections
Phase 1: [answer each question with evidence]
Phase 2: [answer each question with evidence]
Proceeding to Phase 4"
Create the new agent based on gathered context
Required output at the start of this phase:
Working on Phase 4 - Create agent
Required output before proceeding:
"Agent created [path]"
Proceed to Phase 5
Providing report to the user
Required output at the start of this phase:
Working on Phase 5 - Report
The design is done.
File path - [file-path]