360 Secure SkillHub
360 安全技能中心

Team Role Launchkit

提供研究者、构建者、编辑者三种角色模式与启动提示,要求填写交接契约以明确范围与完成标准,并按研究→构建→编辑顺序执行,最终输出交付物、各角色改动、未决风险与下一步建议。适用于复杂、长流程且需角色分离的任务。

cwheeler67
cwheeler67
热度 34.19
最近更新 2026-05-11 17:44:03
分类:自动化依赖:内部调用大模型应用场景:工作流 / 任务管理 / 研究 / 校对润色

安装技能

复制下面的话术发给你的 Agent,它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill: npx secure-skills-cli add team-role-launchkit
Metadata
name
team-role-launchkit
description
Launch clone-like multi-agent workflows using explicit role prompts (researcher, builder, editor), strict handoff contracts, and shared behavior files for consistent long-running execution.

Team Role Launchkit

Use this skill when a single task needs parallelization or role separation without losing coherence.

What this skill provides

  • 3 role modes: Researcher, Builder, Editor
  • A required handoff contract to prevent role blur
  • Launch prompts to start role sessions quickly
  • Guardrails for memory hygiene and approval boundaries

When to use

Use for:

  • complex, multi-step tasks
  • tasks crossing research + implementation + communication
  • long-running tasks where consistency matters

Avoid for:

  • tiny one-shot requests
  • deterministic extraction tasks

Role definitions

  • Researcher: gather facts/options/constraints; no final outward messaging
  • Builder: execute implementation and report exact outputs/diffs
  • Editor: refine final user-facing output without changing core facts

Orchestration sequence

  1. Define objective and constraints.
  2. Fill handoff contract from references/handoff-contract.md.
  3. Run roles in order: Researcher → Builder → Editor.
  4. Operator reviews and delivers final result.
  5. Capture durable outcomes in memory.

Non-negotiables

  • Every role handoff must include done criteria.
  • Receiver must restate scope before acting.
  • External/public actions still require explicit approval policy.
  • Temporary notes are not durable memory unless promoted.

Output contract

Return:

  • final deliverable
  • what changed by role
  • unresolved risks/questions
  • next best action