poorvith-mp/skills-agents

agent-architecture

Design role decomposition, message passing, shared state and failure isolation across agents.

查看源码

使用方法

  1. 复制安装命令,在自己的智能体环境安装前先检查源码包。
  2. 从原始文档核对所需工具、凭据和支持的宿主。
  3. 说明输入材料与期望交付物,并对照来源检查输出。
npx skills add https://github.com/poorvith-mp/skills-agents --skill agent-architecture

开始前需了解

  • 安装不包含模型订阅、API 额度或访问文件的权限。
  • 仓库星数和安装量反映活跃情况,不等于独立质量或安全测试。
仓库原始内容

按源仓库内容呈现,保留标题、案例、代码、表格、链接以及原文引用的演示图片。

Agent Architecture

Design multi-agent systems where coordination does not become the failure mode.

Process

  1. Justify the second agent. Most tasks that look multi-agent are one agent with better tools. Split only when roles need genuinely different context or run concurrently.
  2. Decompose by context boundary, not by job title. An agent exists to hold a distinct working set, not to mirror an org chart.
  3. Define the contract between agents — the exact shape of what is passed, and what is guaranteed about it. Free-text handoff is where multi-agent systems lose information.
  4. Decide where state lives. Shared mutable state between agents needs the same care as between threads.
  5. Isolate failure. One agent failing or looping must not stall the others; set timeouts and a supervisor that can terminate.
  6. Make the trace reconstructable. When output is wrong, you need to see which agent decided what, with what input.
  7. Bound the total cost. Agents that can spawn agents need a hard ceiling.

Deliverables

  • Role decomposition with the context boundary justifying each agent
  • Message contracts between agents
  • Failure isolation, timeouts and supervision model
  • Trace format and cost ceiling

Verification & Quality Checklist

  • [ ] Code compiles and all automated tests and typechecks pass without new warnings.
  • [ ] Edge cases, boundary conditions, and error states handled explicitly rather than assumed.
  • [ ] No hardcoded secrets, credentials, or insecure defaults introduced.
  • [ ] Changes are covered by a test that fails without them.

Anti-Patterns & Constraints

  • NEVER weaken or skip a failing test to make a change land.
  • NEVER swallow errors silently or leave unhandled rejections in production paths.
  • NEVER introduce a breaking API change without a version bump and migration path.