The Agent Skills Directory

[SAFE]: The skill is a configuration guide and security template that helps users align permission models across multiple AI agent platforms. It advocates for the principle of least privilege by scoping automation to specific repository roots and using disposable sandboxes for high-autonomy tasks.
[SAFE]: It provides proactive security controls in the form of 'safety hook' scripts (Bash). These scripts are designed to intercept tool calls in Claude Code (via PreToolUse) and Gemini CLI (via BeforeTool) to programmatically block 'Tier 1' dangerous operations such as recursive deletions, force pushing, database drops, and access to sensitive environment/credential files.
[SAFE]: The instructions explicitly warn against global bypass modes and provide clear guidance on implementing denylists for sensitive paths (e.g., .env, secrets/**) and administrative commands (e.g., sudo). This demonstrates a security-first approach to automation configuration.
[SAFE]: All external references provided in the skill point to official platform documentation or established resources (Anthropic, OpenAI, Google Gemini, agentskills.io). No obfuscated code, hidden instructions, or suspicious remote dependencies were detected.

ralphmode