说明排毒

安全 社区

技能解读

当前原文的自动摘译,尚未人工审校。

目的

使用此技能将说明文件作为行为护栏而不是被动文档进行审查。目标是消除上下文浪费、检测漂移并保持指令加载的可预测性。

何时使用

在以下情况下使用此技能: 指令文件随着时间的推移而积累而没有清理 副驾驶建议感觉不一致、嘈杂或偏离模式 您正在添加新的指令文件并希望避免重叠 您正在检查 applyTo 示波器质量

请勿使用以下情况

在以下情况下不要使用此技能: 调试应用程序运行时问题 实现产品功能 修复与指令质量无关的 lint、测试或构建失败

核心原则

上下文稀缺。每一条指令都必须赢得它的位置。如果指令没有显着改变座席行为,则加强或删除它。

审查程序

清点指令表面并识别规范文件与生成文件。检查每个文件是否有明确的目的和明确的触发点或范围。删除或标记重复的指导、哲学散文和陈旧的规则。审核applyTo模式是否存在过于广泛的匹配和差距。将这些说明与当前的代码库和工作流程现实进行比较。

流狐档案 运行与权限由流狐静态检查推断,不代表作者声明或实际测试结果。
领域
安全
作者声明的智能体
未找到明确声明;不据此推断已兼容或已测试
静态检查
88 / 100 · 启发式扫描,不代表运行安全
运行环境证据
未执行;当前没有隔离运行测试证据
作者 / 版本 / 许可证
@tomevault-io · 未声明许可证
上下文用量评级
低消耗
接入复杂程度
即装即用
是否需要外部接口密钥
不需要
兼容的系统
未声明(默认跨平台)
运行环境要求
未检出明确要求
检测到的文件与系统行为
  • 只读
  • 允许写入 / 修改
网络行为
仅限本地
安装命令数
无(仅作为资料)

需要注意: 未声明工具权限范围;实际权限由运行环境决定。

作者说明文件(原文) SKILL.md

# Instructions Detox Skill

## Purpose

Use this skill to review instruction files as behavioral guard rails rather than passive documentation. The goal is to remove context waste, detect drift, and keep instruction loading predictable.

## When to Use

Use this skill when:

- instruction files have accumulated over time without cleanup
- Copilot suggestions feel inconsistent, noisy, or off-pattern
- you are adding new instruction files and want to avoid overlap
- you are reviewing `applyTo` scope quality
- you need a maintenance audit before merging instruction changes

## Do Not Use When

Do not use this skill when:

- debugging application runtime issues
- implementing product features
- fixing lint, test, or build failures unrelated to instruction quality

## Core Principle

**Context is scarce. Every instruction must earn its place.**

If an instruction does not measurably change agent behavior, tighten it or delete it.

## Review Procedure

1. Inventory the instruction surface and identify canonical versus generated files.
2. Check whether each file has a narrow purpose and a clear trigger or scope.
3. Remove or flag duplicated guidance, philosophical prose, and stale rules.
4. Audit `applyTo` patterns for over-broad matching and gaps.
5. Compare the instructions with the current codebase and workflow reality.
6. Produce a prioritized remediation report with the highest-risk problems first.

## Success Metrics

A healthy instruction ecosystem has:

- small, focused files with one job each
- clear and narrow `applyTo` scopes
- constraints that prevent bad behavior
- minimal duplication across files
- wording that matches actual repository workflows

## Output Contract

Produce a markdown report containing:

1. severity-ordered findings
2. exact file references and line references
3. quoted examples of bloated, stale, or duplicated guidance
4. recommended rewrites, deletions, or scope changes
5. a short action plan for remediation

## Constraints

- Prefer deletion or consolidation over adding more instruction text.
- Do not rewrite generated mirrors before updating the canonical source.
- Do not propose vague style advice without evidence from the files.

## Workflows

1. [Full Detox Review](references/01-full-detox-review.md) - Complete 8-step audit process
2. [Quick Bloat Check](references/02-quick-bloat-check.md) - Fast bloat scan (< 15 min)
3. [Rot Detection](references/03-rot-detection.md) - Verify instructions match reality
4. [Scope Audit](references/04-scope-audit.md) - Review applyTo patterns
5. [Maintenance Review](references/05-maintenance-review.md) - Quarterly check-in

## Agent

Use the **Instructions Detox** agent (`@instructions-detox`) to run these workflows.

---
> Source: [CambridgeMonorail/WallRun](https://github.com/CambridgeMonorail/WallRun) — distributed by [TomeVault](https://tomevault.io).
<!-- tomevault:4.0:skill_md:2026-05-23 -->

讨论

基于 GitHub Discussions。登录 GitHub 即可参与讨论、点赞、订阅更新。