← 返回
未分类 中文

Loop Stability Check

Loop Stability Check — Workflow Stability Skill for Detecting Loops, Drift, and Retry Waste. Use it when the user needs a disciplined protocol and fixed outp...
Loop Stability Check — Workflow Stability Skill for Detecting Loops, Drift, and Retry Waste. Use it when the user needs a disciplined protocol and fixed outp...
clarkchenkai clarkchenkai 来源
未分类 clawhub v1.0.0 1 版本 100000 Key: 无需
★ 0
Stars
📥 351
下载
💾 0
安装
1
版本
#latest

概述

Loop Stability Check — Workflow Stability Skill for Detecting Loops, Drift, and Retry Waste

Use this skill when the task matches the protocol below.

Activation Triggers

  • repeated retries with no improvement
  • oscillation between multiple outputs or states
  • planning loops that never cash out into execution
  • human-bot ping-pong with no stable owner
  • cases where repetition is happening but convergence is not

Core Protocol

Step 1: Define the intended loop objective

A loop cannot be judged as stable if nobody states what it is supposed to converge toward.

Step 2: Observe the repeated behavior

Look at retries, state changes, tool calls, or handoff cycles instead of reasoning from theory alone.

Step 3: Classify the instability

Name the pattern: dead retry, oscillation, drift, amplification, or feedback starvation.

Step 4: Locate the feedback failure

Ask which signal is missing, delayed, noisy, or being ignored.

Step 5: Add guardrails and intervention

Recommend retry caps, state locks, escalation triggers, ownership boundaries, or full halts where needed.

Output Contract

Always end with this six-part structure:

## Loop Objective
[...]

## Observed Behavior
[...]

## Stability Risks
[...]

## Likely Failure Mode
[...]

## Guardrails
[...]

## Recommended Intervention
[...]

Response Style

  • Describe the loop behavior concretely.
  • Prefer named failure modes over generic ‘this seems buggy.’
  • Recommend the smallest guardrail that restores convergence.
  • Stop the loop when the right action is halt, not another cycle.

Boundaries

  • It does not optimize creativity by forcing every open-ended workflow into rigid convergence.
  • It does not confuse ‘more steps’ with ‘more learning.’
  • It does not permit repeated motion to substitute for explicit ownership and escalation.

版本历史

共 1 个版本

  • v1.0.0 当前
    2026-05-07 12:32 安全 安全

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

ai-agent

Agent Browser

rez0
用于 AI 代理的浏览器自动化 CLI。当用户需要与网站交互(包括浏览页面、填写表单、点击按钮、截图等)时使用。
★ 844 📥 324,778
ai-agent

Find Skills

guipi888
场景驱动+关键词双模式技能发现工具。当用户用自然语言描述场景/需求(如"我想做一个海报""帮我分析股票"),或明确说"安装技能/find skills/找个skill"时,自动从官方内置、本地已安装、SkillHub、虾评、GitHub、C
★ 1,492 📥 557,346
ai-agent

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。
★ 1,410 📥 325,137