技能 improve-skill

📦

improve-skill

Name: improve-skill
Author: Dwsy

内容修订 r1 高风险 ⚙️ 外部命令📁 文件系统访问

从智能体会话中改进技能

技能作者需要来自真实智能体会话的清晰证据。此技能会提取转录，并将其转化为用于改进或创建技能的提示词。

支持: Claude Codex Code(CC)

⚠️ 38 差

安装前请停下并请求确认。

请审查安装计划，并在更改文件前获得用户明确同意。

使用我的智能体安装

将此请求复制给你的智能体，其中包含权威 Skill 页面和清单。

智能体请求

Review the Skillstore skill "improve-skill" from https://skillstore.io/skills/dwsy-improve-skill.md and its manifest at https://skillstore.io/api/skills/dwsy-improve-skill/manifest. Verify the artifact. Stop and obtain explicit user consent before installing or changing files.

你的智能体仍应展示其计划，并按安全策略请求必要确认。

Agent 可读资源

当 AI Agent、爬虫或脚本需要干净上下文，而不是读取完整页面时，请使用这些链接。

Markdown 详情 GET /skills/dwsy-improve-skill.md 已签名 manifest GET /api/skills/dwsy-improve-skill/manifest 已签名 lockfile GET /api/skills/dwsy-improve-skill/lockfile

测试它

正在使用“improve-skill”。使用我最新的 Codex 会话改进部署技能。

预期结果:

一份可读的转录摘要、一组技能缺口，以及一个用于在新会话中更新该技能的提示词。

正在使用“improve-skill”。从一次成功的调试会话中创建新技能。

预期结果:

建议的技能重点、关键工作流步骤、常见陷阱和示例用法指导。

正在使用“improve-skill”。审查一次智能体不断重试命令的会话。

预期结果:

对困惑点、缺失示例和面向未来智能体的更清晰说明进行聚焦分析。

安全审计

高风险

v7 • 2026/7/5 打开版本化报告

The shell-execution findings are false positives caused by JavaScript template literals and Markdown backticks. The remaining risk is legitimate filesystem access to hidden agent session and skill directories. Raw transcript reuse also lacks redaction and prompt-injection controls.

已扫描文件

506

分析行数

复核项

已忽略误报

已确认安全问题 (2)

高

SKILL.md:60-88 SKILL.md:101-135

Raw Session Transcript Reuse Can Carry Prompt Injection

The workflow tells users to paste the extracted transcript into a fresh agent session. Untrusted session text can include instructions that steer the new agent unless isolated and reviewed.

The cited prompt templates embed a session transcript into a new agent workflow with no instruction to treat transcript text as untrusted data. This is a common prompt-injection transfer path.

高

scripts/extract-session.js:322-343 SKILL.md:48-50 SKILL.md:94-96

Sensitive Agent Session Data Is Extracted Without Redaction

The script reads local agent session files and prints message content. The documentation also suggests saving the transcript in /tmp, which may expose secrets or private project data.

The script reads the selected session file and prints the formatted transcript, while the documentation redirects output to a temporary file. I found no redaction or secret scanning step in the cited workflow.

能力复核项 (26)

这些是真实的本地能力，对此技能可能是预期行为，因此需要复核，但不会按已确认恶意行为计分。

高

scripts/extract-session.js:322

Node.js fs operations

const content = fs.readFileSync(result.path, 'utf8');

The script reads the selected agent session file and later prints its transcript. Session files can contain sensitive prompts, code, and tool results.

高

SKILL.md:40

Hidden file in home directory

- **Claude Code**: `~/.claude/projects/<encoded-cwd>/*.jsonl`

No specific safe context was identified for this static finding, so it is kept for review with low confidence.

高

SKILL.md:41

Hidden file in home directory

- **Pi**: `~/.pi/agent/sessions/<encoded-cwd>/*.jsonl`

No specific safe context was identified for this static finding, so it is kept for review with low confidence.

高

SKILL.md:42

Hidden file in home directory

- **Codex**: `~/.codex/sessions/YYYY/MM/DD/*.jsonl`

No specific safe context was identified for this static finding, so it is kept for review with low confidence.

高

SKILL.md:54

Hidden file in home directory

- `~/.codex/skills/<skill-name>/SKILL.md`

No specific safe context was identified for this static finding, so it is kept for review with low confidence.

高

SKILL.md:55

Hidden file in home directory

- `~/.claude/skills/<skill-name>/SKILL.md`

No specific safe context was identified for this static finding, so it is kept for review with low confidence.

高

SKILL.md:56

Hidden file in home directory

- `~/.pi/agent/skills/<skill-name>/SKILL.md`

No specific safe context was identified for this static finding, so it is kept for review with low confidence.

高

SKILL.md:118

Hidden file in home directory

Write the skill to: ~/.codex/skills/<skill-name>/SKILL.md

No specific safe context was identified for this static finding, so it is kept for review with low confidence.

中

scripts/extract-session.js:53

Node.js fs operations

const files = fs.readdirSync(dir)

The script enumerates local agent session directories to find transcripts. The behavior is documented, but it accesses sensitive local history.

中

scripts/extract-session.js:77

Node.js fs operations

for (const entry of fs.readdirSync(dir, { withFileTypes: true })) {

The script enumerates local agent session directories to find transcripts. The behavior is documented, but it accesses sensitive local history.

中

scripts/extract-session.js:96

Node.js fs operations

const firstLine = fs.readFileSync(session.path, 'utf8').split('\n')[0];

The script reads Codex session files while searching for the matching working directory. This is local and intended, but it touches sensitive agent history.

中

SKILL.md:40

Hidden file access

- **Claude Code**: `~/.claude/projects/<encoded-cwd>/*.jsonl`

No specific safe context was identified for this static finding, so it is kept for review with low confidence.

中

SKILL.md:41

Hidden file access

- **Pi**: `~/.pi/agent/sessions/<encoded-cwd>/*.jsonl`

No specific safe context was identified for this static finding, so it is kept for review with low confidence.

中

SKILL.md:42

Hidden file access

- **Codex**: `~/.codex/sessions/YYYY/MM/DD/*.jsonl`

No specific safe context was identified for this static finding, so it is kept for review with low confidence.

中

SKILL.md:54

Hidden file access

- `~/.codex/skills/<skill-name>/SKILL.md`

No specific safe context was identified for this static finding, so it is kept for review with low confidence.

中

SKILL.md:55

Hidden file access

- `~/.claude/skills/<skill-name>/SKILL.md`

No specific safe context was identified for this static finding, so it is kept for review with low confidence.

中

SKILL.md:56

Hidden file access

- `~/.pi/agent/skills/<skill-name>/SKILL.md`

No specific safe context was identified for this static finding, so it is kept for review with low confidence.

中

SKILL.md:118

Hidden file access

Write the skill to: ~/.codex/skills/<skill-name>/SKILL.md

No specific safe context was identified for this static finding, so it is kept for review with low confidence.

中

SKILL.md:50

Temp directory access

./scripts/extract-session.js > /tmp/session-transcript.txt

The documentation suggests writing full session transcripts to /tmp. Temporary files can expose sensitive transcript content to other local processes or users.

中

SKILL.md:96

Temp directory access

./scripts/extract-session.js > /tmp/session-transcript.txt

The documentation suggests writing full session transcripts to /tmp. Temporary files can expose sensitive transcript content to other local processes or users.

低

scripts/extract-session.js:51

Node.js fs stat operations

if (!fs.existsSync(dir)) return null;

The script checks local paths and file metadata while locating session transcripts. This is expected for the tool, but it still observes hidden session storage.

低

scripts/extract-session.js:58

Node.js fs stat operations

mtime: fs.statSync(path.join(dir, f)).mtime

The script checks local paths and file metadata while locating session transcripts. This is expected for the tool, but it still observes hidden session storage.

低

scripts/extract-session.js:70

Node.js fs stat operations

if (!fs.existsSync(baseDir)) return null;

The script checks local paths and file metadata while locating session transcripts. This is expected for the tool, but it still observes hidden session storage.

低

scripts/extract-session.js:76

Node.js fs stat operations

if (!fs.existsSync(dir)) return;

The script checks local paths and file metadata while locating session transcripts. This is expected for the tool, but it still observes hidden session storage.

低

scripts/extract-session.js:84

Node.js fs stat operations

mtime: fs.statSync(fullPath).mtime

The script checks local paths and file metadata while locating session transcripts. This is expected for the tool, but it still observes hidden session storage.

低

scripts/extract-session.js:267

Node.js fs stat operations

if (!fs.existsSync(sessionPath)) {

The script checks local paths and file metadata while locating session transcripts. This is expected for the tool, but it still observes hidden session storage.

风险因素

⚙️ 外部命令 (34)

📁 文件系统访问 (28)

审计者: codex 查看审计历史 →

分享与引用此报告

分享版本化评估报告、中性徽章、嵌入卡片和引用。Skillstore 只报告证据，不判断该 Skill 是否安全。

打开版本化报告

复制报告链接

https://skillstore.io/skills/dwsy-improve-skill/audits/7?utm_source=security_passport&utm_medium=share&utm_campaign=versioned_report

Markdown 徽章

[![Skillstore security assessment](https://skillstore.io/badges/skills/dwsy-improve-skill/security.svg)](https://skillstore.io/skills/dwsy-improve-skill?utm_source=security_passport_badge)

HTML 徽章

<a href="https://skillstore.io/skills/dwsy-improve-skill?utm_source=security_passport_badge"><img src="https://skillstore.io/badges/skills/dwsy-improve-skill/security.svg" alt="Skillstore security assessment" loading="lazy"></a>

嵌入卡片

<iframe src="https://skillstore.io/embed/skills/dwsy-improve-skill.html" title="Skillstore Security Assessment" sandbox="allow-popups allow-popups-to-escape-sandbox" loading="lazy" referrerpolicy="no-referrer" width="420" height="180"></iframe>

学术引用 (APA · BibTeX · CFF)

APA 引用

Dwsy. (2026). improve-skill security audit report (audit version 7) [Author version unspecified]. Skillstore. https://skillstore.io/skills/dwsy-improve-skill/audits/7

BibTeX 引用

@techreport{dwsy-dwsy-improve-skill-2026,
  author = {Dwsy},
  title = {improve-skill security audit report (audit version 7)},
  institution = {Skillstore},
  year = {2026},
  number = {7},
  url = {https://skillstore.io/skills/dwsy-improve-skill/audits/7},
  note = {Author version unspecified}
}

CITATION.cff

cff-version: 1.2.0
message: "If you use this Skill, cite its author and this versioned security audit report."
title: "improve-skill security audit report (audit version 7)"
version: "unspecified"
type: report
authors:
  - name: "Dwsy"
date-released: "2026-07-05"
url: "https://skillstore.io/skills/dwsy-improve-skill/audits/7"
identifiers:
  - type: other
    value: "skillstore:dwsy-improve-skill:audit:7"
    description: "Skillstore immutable audit report identifier"

Skillstore 评分

评分依据证据置信度: 中

架构

可维护性

内容

社区

规范符合性

你能构建什么

改进现有技能

审查真实智能体会话，并识别当前技能中缺失的说明、示例和修正。

创建可复用技能

将一次成功的编码会话转化为其他智能体可遵循的可复用工作流。

记录智能体工作流经验

从会话历史中提取重复失败、变通方案和成功模式，供团队审查。

试试这些提示

查找技能缺口

使用 improve-skill 提取我当前的会话，并列出我使用的技能中最重要的缺失说明。

起草改进提示词

分析此项目最新的 Claude Code 会话，并为现有技能起草一份简洁的改进提示词。

使用特定会话

使用这个特定的 Codex 会话文件，识别应添加到我的技能中的示例、修正和警告。

优先处理安全的技能更改

将转录与当前技能进行比较，并生成包含遮蔽说明和风险检查的优先级编辑计划。

最佳实践

在与另一个智能体共享转录之前，先审查其中是否包含密钥。
使用显式技能路径，并在编辑前确认目标技能。
让改进提示词聚焦于观察到的失败和成功模式。

避免

不要将原始生产密钥或私有客户数据粘贴到新会话中。
未经同意，不要在共享机器上运行转录提取。
不要将生成的改进提示词视为已经审查过的技能更改。

常见问题

此技能支持哪些会话格式？

它支持该技能文档中记录的 Claude Code、Pi 和 Codex JSONL 会话文件。

它会自动编辑技能吗？

不会。它帮助生成提示词和工作流，用于指导单独的技能编辑会话。

它可以使用特定的会话文件吗？

可以。当自动检测不够时，提取器接受显式会话路径。

它会遮蔽敏感数据吗？

不会。用户应在将转录复制到另一个智能体会话之前进行审查和遮蔽。

为什么它使用新的会话？

新的会话可以让改进任务聚焦于转录和目标技能。

它适用于私有仓库吗？

请谨慎使用。会话转录可能包含私有代码、路径、命令和工具输出。

开发者详情

作者

Dwsy

许可证

MIT

Skillstore 修订

版本提示

作者未声明版本。

仓库

https://github.com/Dwsy/agent/tree/main/skills/improve-skill

引用

c579efd02184f1128d6741d16e6c8d951d7d4208

维护新鲜度

2026/7/20

使用量

6 次下载 · 217 次浏览

文件结构

📁 scripts/

📄 extract-session.js

创建可复用的 AI 技能

由 vercel-labs

团队需要一致的方法来打包可复用的智能体知识和工作流。此技能会指导 Claude、Codex 和 Claude Code 完成技能规划、编写、验证和打包。

编码与开发查看

improve-skill

使用我的智能体安装

Agent 可读资源

测试它

安全审计

已确认安全问题 (2)

风险因素

复制报告链接

Markdown 徽章

HTML 徽章

嵌入卡片

APA 引用

BibTeX 引用

CITATION.cff

Skillstore 评分

你能构建什么

改进现有技能

创建可复用技能

记录智能体工作流经验

试试这些提示

最佳实践

避免

常见问题

开发者详情

Dwsy 的更多技能

project-planner

svg-logo-generator

context7

system-design

ast-grep

ai-runtime-memory

相关技能

improving-components

building-agents

agent-development

building-commands

analyzing-component-quality

skill-creator