360 Secure SkillHub
360 安全技能中心

Skillhub Publish Auditor

该 Skill 以公开包的标准审计待发布的 Agent Skill 文件夹，按结构、发现性元数据、脚本安全和打包四道关卡检查，运行内置 Python 审计脚本并输出分级问题清单与文件级修复建议，最终给出 READY、READY_WITH_WARNINGS 或 BLOCKED 的发布结论。

debtvc2022
debtvc2022
热度 23.22
最近更新 2026-06-18 17:01:20
分类：开发依赖：代码执行 / 读写文件应用场景：代码审查 / DevOps / 报告撰写 / 工作流

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 skillhub-publish-auditor
Metadata
name
skillhub-publish-auditor
description
Use when preparing a Skill folder for public release to SkillHub, ClawHub, or another agent-skill marketplace, especially before publishing third-party installable skills, scripts, references, metadata, or examples.

SkillHub Publish Auditor

Overview

Audit an agent Skill as a public package, not a private note. The goal is to catch release blockers, weak discovery metadata, unsafe scripts, credential leaks, unfinished template markers, and packaging issues before upload.

Quick Start

Run the bundled auditor from the skill folder you want to publish:

python <this-skill>/scripts/audit_skill_release.py <skill-folder> --markdown

Use --json for CI or automated marketplace checks.

Release Standard

Apply four gates in order:

  1. Structure: SKILL.md exists, frontmatter parses, name is publishable, and no unfinished template markers remain.
  2. Discovery: description starts with Use when, names concrete triggers, and does not summarize the whole workflow.
  3. Safety: scripts and references do not read secrets, hide instruction-override payloads, run destructive shell commands, or send local data out.
  4. Packaging: resource links resolve, examples are intentional, auxiliary files are not clutter, and UI metadata is useful.

Read references/release-checklist.md when judging borderline findings or preparing a manual review.

Required Behavior

  • Treat every file in the candidate skill as publishable supply-chain content.
  • Inspect scripts before recommending release, even when SKILL.md looks clean.
  • Mark issues as blocker, warn, or info; keep each finding specific.
  • Recommend exact file-level fixes, but do not auto-edit unless the user asks.
  • If the skill executes code, require at least one local smoke test or explain why no test can run.

Common Mistakes

Mistake Fix
Publishing template text Remove unfinished markers and generated instructions.
Description explains workflow Rewrite it as trigger conditions only.
Security review stops at SKILL.md Scan every script, reference, config, and example.
Marketplace metadata ignored Check agents/openai.yaml when present.
"Harmless" helper scripts read env files Flag environment, SSH, credential, token, and key access.

Output Contract

End with a release decision:

  • READY: no blockers.
  • READY_WITH_WARNINGS: warnings remain, but no release blocker.
  • BLOCKED: one or more blockers must be fixed before publishing.