Token导航 LogoToken导航

Agent Skill · 研究检索

auto-context(自动上下文)

auto-context 用于查找、检索和筛选相关信息,适合在 OpenClaw 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。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。

auto-context 是研究检索类 Agent Skill,可用宿主为 OpenClaw,通过ClawHub获取,权限为只读;许可证为 MIT-0,第三方扫描已通过,按安装量在研究检索类的 73,675 个 Skill 里排第 3,913。

总安装 38,202周安装 1,641GitHub Stars 2下载 13,391核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
MIT-0
版本
1.0.0

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:auto-context(自动上下文)
来源仓库:https://github.com/wpank/auto-context
安装命令:
openclaw skills install auto-context
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 auto-context

这个 Skill 是做什么的

研究检索类 Skill 主要解决「找资料、核对事实、把来源整理清楚」的问题,通常会让助手先检索再总结。

在采取重大行动之前自动阅读相关上下文。加载 TODO.md、roadmap.md、交接、任务计划和其他项目上下文文件,以便 AI 在完全态势感知的情况下运行。在启动任务、实现功能、重构、调试、规划或恢复会话时使用。

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 OpenClaw;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 openclaw skills install auto-context;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它标注为只读,先用只读任务试一次,确认行为符合预期再扩大使用范围。

常见问题

6 条

auto-context 怎么安装?

在终端执行 openclaw skills install auto-context。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

auto-context 支持哪些 AI 工具?

公开资料标注支持 OpenClaw。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

auto-context 需要什么权限?

权限级别是只读。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。

auto-context 是免费的吗?

来源标注的许可证是 MIT-0。本站不代收任何费用,也不托管安装包;能否商用、要不要署名,以许可证原文为准。

auto-context 用的人多吗?

公开数据是累计安装 38,202 次,近一周安装 1,641 次,下载 13,391 次,GitHub 2 星。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

auto-context 安全吗?

第三方扫描结果为通过,但这只覆盖已知的静态检查项,不能替代自己读一遍源码和安装命令。

值得选的地方和要先想清楚的地方

值得选的地方

  • 专注 OpenClaw 一种宿主,触发方式和使用习惯不用来回切换。
  • 许可证明确为 MIT-0,商用和分发边界好判断。
  • 第三方扫描已通过,常见的高风险行为没有命中。
  • 标注为只读权限,试用阶段不容易误改本地文件。

先想清楚的地方

  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 OpenClaw,想直接扩展它能力的人。
  • 需要处理研究检索类任务、又不想每次都手写提示词的人。
  • 需要许可证清楚、并且有第三方扫描结果的团队项目。

不太适合

  • 需要厂商级技术支持和 SLA 的生产系统。

研究检索类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
find-skills

寻找技能 · Codex、Claude

31,199,81316,486操作浏览器MIT有提醒
self-improving-agent

自我改进 Agent · OpenClaw

16,464,0023,396接触敏感数据MIT-0通过
soultrace

灵魂追踪 · Codex、Claude

7,588,94412需要联网未明确有提醒
self-improving

自我完善 · OpenClaw

6,829,2841,046需要联网MIT-0通过
microsoft-foundry

微软代工厂 · Codex、Claude

6,614,144764调用外部服务未明确有提醒
auto-context

自动上下文 · 本页

38,2022只读MIT-0通过

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

name
auto-context
model
fast
description
Automatically read relevant context before major actions. Loads TODO.md, roadmap.md, handoffs, task plans, and other project context files so the AI operates with full situational awareness. Use when starting a task, implementing a feature, refactoring, debugging, planning, or resuming a session.

Auto-Context — Situational Awareness Protocol (Meta-Skill)

Before you act, understand where you are. This skill ensures the AI loads critical project context automatically before any major action, preventing wasted effort, duplicate work, and misaligned implementations.

Installation

OpenClaw / Moltbot / Clawbot

npx clawhub@latest install auto-context

When to Activate

This skill triggers automatically based on the current action. Do not wait for the user to ask — proactively load context when any of these conditions are met.

TriggerWhyMinimum Context
Starting a new taskUnderstand priorities, avoid conflictsCritical + High
Implementing a featureKnow the plan, constraints, recent changesCritical + High
Refactoring codeUnderstand what changed recently, what's plannedCritical + High + Medium
Debugging an issueCheck recent changes, known issues, discoveriesCritical + High + Medium
Planning or scoping workFull picture of roadmap, backlog, progressAll levels
Session start or resumeRebuild mental model from last session stateCritical + High
Before a handoffEnsure nothing is missed in transitionAll levels

Context Files to Read

Read these files in priority order. Stop early if the task is narrow and lower-priority files are clearly irrelevant.

PriorityFilePurposeRead When
CriticalTODO.mdCurrent tasks, backlog, and prioritiesAlways
Criticalroadmap.mdPhase status, milestones, project directionAlways
Hightask_plan.mdActive task breakdown and implementation planFile exists
High.cursor/handoffs/*.mdRecent handoff notes (read last 3 by date)File exists
Mediumfindings.mdResearch results, discoveries, decisions madeRelevant to task
MediumCHANGELOG.mdRecent changes and their rationaleRelevant to task
Low.cursor/sessions/*.mdSession summaries (read last 2 by date)Planning or debugging

Alternate Locations

Some projects use different paths. Check these fallbacks if primary paths are empty:

PrimaryFallback
TODO.mddocs/TODO.md, ai/TODO.md
roadmap.mddocs/roadmap.md, ROADMAP.md
task_plan.mddocs/task_plan.md, .cursor/task_plan.md
findings.mddocs/findings.md, .cursor/findings.md

Context Loading Strategy

Step 1: Load Critical Files (Always)

Read TODO.md → Extract: current task, next priorities, blockers
Read roadmap.md → Extract: current phase, active milestone, upcoming deadlines

If either critical file is missing, warn the user:

"No TODO.md found. Consider creating one to track tasks."

Step 2: Load High-Priority Files (If They Exist)

Read task_plan.md → Extract: implementation steps, acceptance criteria
Glob .cursor/handoffs/*.md → Read last 3 by modification date

Step 3: Load Medium/Low Files (If Relevant)

Only read these when the current task benefits from historical context:

  • Debugging? — Read findings.md and CHANGELOG.md
  • Planning? — Read everything including session files
  • Quick fix? — Skip medium and low entirely

Step 4: Synthesize and Present

After loading, produce a context summary (see format below) before proceeding with the task.


Staleness Detection

Check modification dates on all loaded files. Flag files that may contain outdated information.

AgeStatusAction
< 24 hoursFreshUse as-is
1-7 daysCurrentUse as-is, note the age
7-30 daysStaleWarn: "{file} last updated {N} days ago — verify before relying on it"
> 30 daysOutdatedWarn: "{file} is {N} days old and may no longer reflect project state"

To check file ages on macOS:

stat -f "%m %N" TODO.md roadmap.md task_plan.md findings.md CHANGELOG.md 2>/dev/null

On Linux:

stat -c "%Y %n" TODO.md roadmap.md task_plan.md findings.md CHANGELOG.md 2>/dev/null

Context Summary Format

After loading context, present a concise summary using this template. Keep it tight — the goal is awareness, not repetition.

## Context Loaded

**Current Phase:** {phase from roadmap}
**Active Milestone:** {milestone and progress}

**Current Task:** {from TODO.md or task_plan.md}
- Status: {in-progress / blocked / not started}
- Blockers: {any blockers, or "none"}

**Recent Changes:**
- {last 2-3 items from CHANGELOG or handoffs}

**Relevant Findings:**
- {key discoveries that affect the current task, or "none"}

**Stale Warnings:**
- {any staleness warnings, or "all context is fresh"}

If no context files exist at all, output:

## Context Loaded

No project context files found. Operating without historical context.
Consider creating TODO.md and roadmap.md to enable context-aware assistance.

Integration Points

This skill connects to other workflow commands and should run as a precursor.

CommandHow Auto-Context Integrates
/start-taskLoads full context before beginning work; populates task plan
/intentReads roadmap and TODO to validate intent against project direction
/workflowProvides the "understand" phase of any workflow automatically
/progressUses TODO.md and task_plan.md to assess completion status
/handoff-and-resumeReads last handoff to rebuild state on resume
/session-summaryCross-references loaded context with session actions for accuracy

Execution Order

User triggers action
  → Auto-Context activates (this skill)
    → Context summary presented
      → Primary skill/command executes with full awareness

Quick Reference

Context Loading Checklist:
  1. Read TODO.md (critical)
  2. Read roadmap.md (critical)
  3. Read task_plan.md (if exists)
  4. Read last 3 handoffs (if exist)
  5. Check file staleness
  6. Read findings/changelog (if relevant)
  7. Present context summary
  8. Proceed with task

NEVER Do

  1. NEVER skip critical files to save time — leads to duplicate work and conflicting implementations
  2. NEVER load every file regardless of task — wastes tokens and dilutes focus with irrelevant info
  3. NEVER ignore staleness warnings — stale context causes decisions based on outdated assumptions
  4. NEVER read files without summarizing — raw file dumps overwhelm; always synthesize first
  5. NEVER assume context from memory alone — sessions are stateless; always re-read files
  6. NEVER silently proceed when no context exists — user should know they're operating blind
  7. NEVER read handoffs/sessions beyond the limit — last 3 handoffs and last 2 sessions are sufficient

平台分布

OpenClaw

77.92%
按下载量换算10,434

安全审计

VirusTotal

通过

ClawScan

通过

Static analysis

未展示

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。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills