Token导航 LogoToken导航TokenDH.com
开发只读clawhub未标认证来源可访问clear审计通过

autoreason-lite自动推理精简版

Agent Skill

autoreason-lite 用于补充开发相关能力,适合在 OpenClaw 中需要让 Agent 承接开发相关任务时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

总安装

2,472

周安装

101

GitHub Stars

公开资料未说明

下载量

792
OpenClaw

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

MIT-0

最后核验

2026-05-01

来源状态

来源可访问

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:autoreason-lite(自动推理精简版)
来源仓库:https://github.com/cwheeler67/autoreason-lite
安装命令:
openclaw skills install autoreason-lite
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 OpenClaw 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

ClawHubOpenClaw
openclaw skills install autoreason-lite

简介

应用有界多候选者自我完善循环改进草案与计划。

  • 包含 A/B/AB 测试与法官选项防止范围蔓延。
  • 适合分析、草案撰写与复杂推理类开发任务。
  • 安装命令:openclaw skills install autoreason-lite。
  • 应限制迭代次数以避免资源过度消耗。autoreason-lite 属于开发类 Skill,可作为该场景下的辅助能力补充。

SKILL.md

name
autoreason-lite
description
Apply a bounded multi-candidate self-refinement loop (A/B/AB + judges + do-nothing option) to improve drafts, plans, and analyses while preventing scope creep. Use when user asks to improve writing/strategy/code explanations iteratively with quality control.

Autoreason Lite

Use this skill when a user asks to "improve this," "refine this," or "make this better" and quality matters more than one-shot speed.

What this does

A bounded refinement tournament:

  1. A (incumbent): current draft unchanged
  2. B (adversarial revision): deliberately different attempt
  3. AB (synthesis): merge best parts of A and B
  4. Judging pass: pick winner with explicit rubric
  5. Convergence rule: stop early if A keeps winning (no-change is valid)

This reduces over-editing and drift.

Default operating profile

  • Max rounds: 3
  • Judges: 3 independent judge personas
  • Aggregation: Borda-like ranking (1st=2 pts, 2nd=1 pt, 3rd=0)
  • Convergence: stop if A wins 2 rounds (or winner unchanged 2 rounds)
  • Length guardrail: output within ±15% of requested length unless user asks otherwise
  • Voice lock: preserve user's tone profile (technical / founder / viral) unless asked to shift

When to use

Use for:

  • Long-form writing
  • Strategy memos
  • Explanations/tutorial drafts
  • Product copy
  • Decision frameworks

Avoid for:

  • Deterministic factual extraction
  • Tiny edits user already specified exactly
  • Time-critical one-liners unless user requests deep refinement

Execution steps

  1. Clarify success criteria (tone, audience, length, goal) if missing.
  2. Generate candidate B from A:

- Must change structure or argument order (not just wording). - Must preserve critical facts.

  1. Generate candidate AB:

- Keep strongest parts from A and B. - Remove redundancy.

  1. Run 3 judges independently with rubric:

- Accuracy / faithfulness - Clarity - Usefulness for user goal - Concision / scope control

  1. Score candidates; choose winner.
  2. Repeat up to max rounds with winner as new A.
  3. Return final with short “what changed + why”.

Judge prompt template

Use references/judge-rubric.md.

Output format to user

  • Final refined result
  • 2-4 bullets: key improvements
  • Optional: one-line note if loop stopped due to convergence (no meaningful gain)

Quick presets

  • Technical: precise wording, fewer claims, concrete mechanisms
  • Founder: outcomes + positioning + credibility signal
  • Viral: short lines, strong hooks, high readability, no fluff

Safety + quality constraints

  • Never invent facts to make prose sound better.
  • Keep user intent stable unless explicitly asked to pivot.
  • Prefer no-change over noisy edits.
  • If confidence drops, surface uncertainty instead of bluffing.

适合场景

01

OpenClaw 用户查找和安装 Skill 时

02

用户想查找某类 Agent Skill 时

03

需要根据任务场景推荐可安装能力包时

04

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

补充不同宿主或平台的使用分布数据

能力 5

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

OpenClaw

93.96%
按下载量换算744

安全审计

VirusTotal

通过

ClawScan

通过

Static analysis

通过

权限和风险

只读

该 Skill 主要提供规则、说明或参考内容,本身偏只读;真正读写文件、联网或执行命令仍取决于宿主 Agent 的任务。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills