Install Governance

Install Governance

Skill installation is a small supply-chain decision. The curator should treat it as reviewable and reversible.

Default Policy

• Do not silently install unknown third-party skills.
• Show candidates, scores, and risks before installing.
• Install only the selected skill folder when possible.
• Do not bulk-install large awesome lists.
• Preserve a timestamped backup if a destination already exists.
• After install, read the installed SKILL.md and verify it still matches the current task.

Candidate Presentation

Before installation, present:

Field	Example
Repository	owner/repo
Skill path	path/to/skill
Why it matches	Task fit and expected reuse
Repository signals	Stars, forks, update date, license
Score and tier	Strict, relaxed, exploratory
Safety notes	Risks and unknowns
Install target	Codex, Claude Code, or both

Post-Install Behavior

When the user approves installing a skill, treat that approval as intent to use the skill for the current task unless the user explicitly asked only to install it.

After successful install:

1. Read the installed SKILL.md.
2. Confirm name, description, and body match the task.
3. Apply it to the user's original task if safe.
4. Print future invocation syntax.

Invocation examples:

Codex: $skill-name ...
Claude Code: /skill-name ...

When Not to Invoke

Do not invoke a newly installed skill if it proves mismatched, unsafe, overbroad, or impossible to use in the current environment. Explain the reason and suggest disable, quarantine, or removal.

中文说明

Skill 安装是一种小型供应链决策，必须可审计、可撤销、可解释。

默认策略：

• 不静默安装未知第三方 skill；
• 安装前展示候选、分数和风险；
• 尽量只安装选中的 skill folder；
• 不批量安装大型 awesome list；
• 目标路径已存在时保留带时间戳的备份；
• 安装后读取已安装的 SKILL.md，确认它仍然匹配当前任务。

用户同意“查看候选”不等于同意“安装”。用户同意“安装 skill”也不等于允许读取密钥、执行破坏性命令或覆盖用户原始任务。