跳到正文
cccskills
免费在 GitHub 公开

super-agent

Use as the single visible starter and controller for workspace work that may change files, runtime assets, planning state, docs, settings, or governance. Normalize the request into the Super Agent lifecycle: intake, spec registration for non-trivial work, planning registration, specialist routing, execution, testing, review, closeout, and planning-state updates before implementation begins.

中文简介正在准备中。当前显示原文说明。

查看安装方法

包含的文件(1)

  • SKILL.md1.9 KB

SKILL.md(原文)

安装前可以查看交给 AI 代理的指令内容。

Super Agent

Load context first

  1. .github/AGENTS.md
  2. .github/copilot-instructions.md
  3. .github/instructions/super-agent.instructions.md

Claude-native execution

  • Use EnterPlanMode before spec and planning stages.
  • Spawn Plan agent for spec registration (planning/specs/active/).
  • Spawn Plan agent for planning registration (planning/active/).
  • Spawn Explore agent for context pack assembly when multi-domain.
  • Spawn general-purpose agent for implementation, testing, and review.
  • Surface activation banner once at the start of the first substantive reply.

Responsibilities

  • Normalize the request before any planning or implementation.
  • Decide whether work is trivial, change-bearing, or non-trivial design-bearing.
  • Ask up to 3 concise clarification questions and stop before planning when ambiguity would materially change scope, architecture, runtime behavior, validation, or safety.
  • Enforce the lifecycle order; do not skip stages.
  • Always produce a commit message suggestion at closeout.
  • Always update planning state before claiming completion.

Output contract

  1. Normalized request summary
  2. Clarification questions first when required by material ambiguity
  3. Plan decision and active plan path
  4. Selected specialist chain (Claude agent types)
  5. Execution mode (sequential or parallel-safe)
  6. Validation obligations
  7. Closeout obligations
  8. Planning update instructions
  9. Activation banner (once, near start of first reply)
  10. Agents used: line in terminal completion

评价

还没有评价。欢迎分享你的使用感受。

评论

无需登录即可发表。欢迎分享使用感受、提问,或说明在哪些环境下无法运行。

还没有评论。来写第一条吧。

提交后会立即公开。请勿填写邮箱、电话等个人信息。URL 最多 2 个。

0 / 1000 字

此表单受 reCAPTCHA 保护,适用 Google 的隐私权政策和服务条款。

同一仓库的技能

Use for non-trivial feature, behavior, workflow, or architecture work before execution planning. Create or update a versioned spec under planning/specs/active with design intent, alternatives, risks, acceptance criteria, and planning readiness.

中文简介正在准备中。当前显示原文说明。

ThiagoGuislotti/copilot-instructions22026年6月29日 更新

Use for non-trivial, design-bearing, or architecture-affecting work before execution planning. Creates or updates a versioned spec under planning/specs/active/ with design intent, alternatives, risks, and acceptance criteria.

中文简介正在准备中。当前显示原文说明。

ThiagoGuislotti/copilot-instructions22026年6月29日 更新

Build a minimal context pack and recommend the correct specialist path for token-efficient execution. Use when the task is non-trivial, spans multiple domains, or should reduce token usage before implementation.

中文简介正在准备中。当前显示原文说明。

ThiagoGuislotti/copilot-instructions22026年6月29日 更新

Assembles the minimal context pack for a given task. Use when the task spans multiple domains or the context set has obvious redundancy. Do not trim required working context by default.

中文简介正在准备中。当前显示原文说明。

ThiagoGuislotti/copilot-instructions22026年6月29日 更新

Route tasks using this repository's static routing catalog and load only the minimal context pack.

中文简介正在准备中。当前显示原文说明。

ThiagoGuislotti/copilot-instructions22026年6月29日 更新

Route tasks using the repository's static routing catalog and load only the minimal context pack. Use before execution when the task spans multiple domains or the correct instruction set is unclear.

中文简介正在准备中。当前显示原文说明。

ThiagoGuislotti/copilot-instructions22026年6月29日 更新

查看 ThiagoGuislotti 的全部技能

报告此技能的问题