← 返回
AI智能 中文

Socrates

Use when an agent is about to plan, choose an approach, do non-trivial coding, refactor, decompose work, or delegate and the quality of execution depends on...
用于代理计划、选方案、做复杂编码、重构、分解工作或委托,执行质量取决于…
hongyi3
AI智能 clawhub v0.1.1 2 版本 100000 Key: 无需
★ 1
Stars
📥 593
下载
💾 54
安装
2
版本
#latest

概述

Socrates

Use this skill to think before acting. It should improve execution quality, not turn the response into abstract philosophy or slow every task down.

Quick Trigger Guide

Use Socrates for:

  • plan creation and sequence design
  • architecture choices or risky implementation decisions
  • refactors with a non-trivial blast radius
  • delegation to one or more sub-agents

Skip or compress Socrates for:

  • trivial chat
  • direct translation
  • formatting-only edits
  • fully specified micro-fixes with almost no rework risk

Core Protocol

  1. Decide whether to skip, compress, or run the full pass.
  2. Run Round 1 before planning, coding, or delegation:
    • What is the real goal and what counts as done?
    • What is known, what is unknown, and which assumptions am I making?
    • What is most likely to fail, waste time, or force rework?
    • Which approach is best, and why am I rejecting the main alternatives?
    • What should happen next, and what should be delegated?
  3. Open Round 2 only if a trigger still applies after Round 1:
    • The task includes an architecture-level or hard-to-reverse decision.
    • Key constraints are missing and acting now is likely to cause rework.
    • Multiple sub-agents will be involved and boundaries are still fuzzy.
    • Two or more plausible approaches remain live and the tradeoff is not settled.
  4. Keep Round 2 narrow. Ask only the unresolved follow-up questions; do not repeat the full pass.
  5. Produce the final sections in order:
    • Socratic Pass
    • Action Plan
    • Delegation Contract only when sub-agents are involved

Compression Rules

  • Skip the full pass for trivial chat, direct translation, formatting-only edits, or fully specified micro-fixes.
  • Use a compressed one-round pass for simple but still important work.
  • Keep the default experience short. Most tasks should finish in one round.

Delegation Rules

  • Run the parent pass before spawning sub-agents.
  • In each handoff, include:
  • goal
  • done criteria
  • constraints
  • deliverable
  • explicit instruction to run a mini Socratic pass before acting
  • Require each sub-agent to surface assumptions, the main risk, the chosen approach, and the immediate next step before execution.

Guardrails

  • Do not write abstract philosophy.
  • Do not restate the user request at length.
  • Do not force a second round by default.
  • Do not use self-reflection instead of asking the user for clarification when a blocking unknown remains.

References

  • Read references/protocol.md for the detailed protocol, templates, and delegation contract.
  • Read references/examples.md for planning, coding, and delegation examples.

版本历史

共 2 个版本

  • v0.1.1 当前
    2026-03-31 15:24 安全 安全
  • v0.1.0
    2026-03-14 04:18

安全检测

腾讯云安全 (Keen)

安全,无风险
查看报告

腾讯云安全 (Sanbu)

安全,无风险
查看报告

🔗 相关推荐

developer-tools

Asking Until 100

hongyi3
OpenClaw 的上下文感知提问协议,在执行编码、项目构建、架构设计、调试和实现前增加澄清步骤。
★ 1 📥 490
ai-intelligence

Self-Improving + Proactive Agent

ivangdavila
自我反思+自我批评+自我学习+自组织记忆。智能体评估自身工作、发现错误并持续改进。
★ 1,360 📥 318,616
ai-intelligence

Proactive Agent

halthelobster
将AI智能体从任务执行者升级为主动预判需求、持续优化的智能伙伴。集成WAL协议、工作缓冲区、自主定时任务及实战验证模式。Hal Stack核心组件 🦞
★ 836 📥 213,248