360 Secure SkillHub
360 安全技能中心

Skill Publish Vetter

面向 ClawHub Skill 发布流程的隐私扫描工具：运行 scripts/publish-check.sh 检测令牌、API Key、密码、私钥、.env 值、个人邮箱与内网 IP，按 CRITICAL/WARNING/INFO 分级输出完整报告，并强制要求用户二次确认后才执行发布。

weidongkl
weidongkl
热度 47.61
最近更新 2026-05-11 17:46:05
分类：安全依赖：代码执行 / 读写文件应用场景：代码审查 / 工作流 / DevOps

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add skill-publish-vetter
Metadata
name
skill-publish-vetter
version
1.1.0
description
Pre-publish privacy scan for ClawHub skills. Detects tokens, keys, credentials, .env secrets, personal info, and internal IPs before publishing.

Skill Publish Vetter 🛡️

Pre-publish privacy scan for ClawHub skills.

Every skill must pass this scan before publishing to ClawHub.

When to Use

  • User says "publish skill to ClawHub"
  • User says "upload skill to ClawHub"
  • User says "update skill on ClawHub"
  • Any request to submit a skill to a public registry

Core Principle

Publishing to ClawHub = publicly visible. Any leaked secret is exposed to the world.

Better safe than sorry. Even if the scan passes, always ask for user confirmation before publishing.

Publish Workflow (Mandatory)

When the user requests a publish/upload/update to ClawHub, follow these steps strictly:

Step 1: Identify the Target Skill

  • If the context clearly identifies the target skill, proceed.
  • Otherwise, ask the user which skill to publish.

Step 2: Run the Privacy Scan

The scan script lives in this skill's scripts/publish-check.sh. Run it from this skill's directory:

bash scripts/publish-check.sh <target-skill-directory>

It scans for:

Category Detection Pattern Example
Tokens / Keys sk-, ghp_, github_pat_, gpg_, xoxb-, glpat-, JWT tokens sk-abc123...
API Keys api_key, apikey, API_KEY, access_token with actual values API_KEY=abc123
Passwords password, passwd, pwd with non-placeholder values password=mysecret123
Private Keys BEGIN.*PRIVATE KEY PEM private key
Public Keys BEGIN.*PUBLIC KEY, ssh-rsa long strings SSH public key
.env Values .env files with actual (non-placeholder) assignments .env: DB_PASS=real_password
Hardcoded Creds Authorization: Bearer with real tokens curl -H "Authorization: Bearer sk-xxx"
Personal Emails Non-placeholder, non-org emails zhangsan@gmail.com
Personal Paths /home/username/, /Users/username/ absolute paths /Users/weidongkl/.ssh/id_rsa
Internal IPs 192.168.x.x, 10.x.x.x, 172.16-31.x.x https://192.168.1.100:8080

Step 3: Output the Full Report

Output the complete scan report to the chat. Do not summarize or omit anything.

Step 4: Secondary Confirmation (Required)

Always ask for confirmation before publishing — even if the scan is clean.

Confirmation message must include:

  1. Skill name
  2. Skill directory path
  3. Full scan report
  4. A clear confirmation prompt

Template:

📋 Publish Confirmation

Skill: <name>
Path: <directory>
Version: <version>
Scan Result: <PASS / ISSUES FOUND>

[Full report here]

Reply "yes" or "confirm" to proceed with publishing, or "cancel" to abort.

Step 5: Wait for User Response

  • "yes" / "confirm" / "ok" / "go" → proceed to Step 6.
  • "cancel" / "no" / "stop" → abort. Do nothing.
  • No response → do not publish. Wait.

Never skip confirmation. Never auto-publish.

Step 6: Publish

After confirmation:

clawhub publish <skill-directory> --slug <slug> --name "<name>" --version "<version>" --changelog "<changelog>"

Ask the user for slug, name, version, and changelog if not provided.

Step 7: Report Result

Tell the user whether publishing succeeded or failed.

Risk Levels

Level Meaning Action
🚨 CRITICAL Token, key, password, private key with actual values Block publish. User must fix first.
⚠️ WARNING Personal email, personal path, internal IP Recommend fixing before publishing.
💬 INFO author, repository identity fields, metadata env exposure Ask user if intentionally public.

Red Lines (Auto-Block)

If any of these are found, refuse to publish by default:

  1. Any token with actual values (Bearer tokens, API keys, Access tokens)
  2. Any private key content (PEM format, SSH private keys)
  3. Any password/credential with actual values (not placeholders)
  4. .env files with actual configuration values
  5. Hardcoded internal IPs or domains
  6. Base64-encoded sensitive data

If the user explicitly says "publish anyway despite risks", re-confirm once before proceeding.

Placeholder Reference

Use these placeholders when fixing issues:

Type Placeholder
Token your-api-token / <YOUR_TOKEN>
API Key your-api-key / <API_KEY>
Password your-password / <PASSWORD>
Email you@example.com / <YOUR_EMAIL>
Username your-username / <USERNAME>
URL https://your-server.example.com
IP your-server-ip

Scan Script

The script is at scripts/publish-check.sh relative to this skill's directory. It uses no absolute paths and works in any installation location.

bash scripts/publish-check.sh <target-skill-directory>

Safety first, publishing second. Never publish without confirmation. 🛡️