Can it automatically fix my prompt files?

Yes, by using the --fix flag, the skill can automatically correct high-certainty issues like missing frontmatter, unrestricted Bash access, and weak constraint language.

How does this help with agent security?

It identifies over-privileged tool configurations, such as unrestricted Bash access, and suggests scoped alternatives like Bash(git:*) to prevent unauthorized command execution.

How does this improve LLM instruction following?

It reorganizes instructions to place critical identity and constraints in the 'high-recall' zones (the beginning and end of the prompt) and replaces vague language with definitive constraints like MUST and NEVER.

What platforms are supported by this skill?

It supports agent prompt files specifically designed for Claude Code, OpenCode, and Codex platforms, looking for platform-specific configuration files.

What is the 'Intern Test' for agent descriptions?

The Intern Test is a validation check to ensure an agent's description is clear enough that a human (or an LLM router) can correctly decide exactly when to invoke the agent based on its stated triggers.

Agent Prompt Optimizer

Name: Agent Prompt Optimizer
Author: avifenesh

byavifenesh

•

311

開発者ツール

Audits and enhances AI agent prompt files using prompt engineering best practices and automated fixes.

概要

This skill provides a comprehensive auditing framework for Claude Code, OpenCode, and Codex agent files, ensuring they adhere to high-performance prompt engineering standards. It evaluates frontmatter, tool restrictions, instructional clarity, and structural layout to minimize hallucination and maximize tool-use reliability. By performing 30 distinct pattern checks and offering automated fixes, it transforms vague agent definitions into precise, production-ready AI personas with proper security boundaries and optimized instruction ordering.

主な機能

Structural optimization focusing on LLM attention and recall patterns
Enforcement of the Principle of Least Privilege for security and performance
Automated analysis of agent frontmatter and tool restrictions
Intern Test validation for clear and actionable agent descriptions
Auto-correction for weak constraint language and missing role definitions
311 GitHub stars

ユースケース

Refining custom Claude Code agents for production-grade reliability and task success
Auditing legacy prompt files to migrate between Claude Code, OpenCode, and Codex
Securing AI agents by automatically identifying and restricting broad Bash tool access

概要

主な機能

Structural optimization focusing on LLM attention and recall patterns
Enforcement of the Principle of Least Privilege for security and performance
Automated analysis of agent frontmatter and tool restrictions
Intern Test validation for clear and actionable agent descriptions
Auto-correction for weak constraint language and missing role definitions
311 GitHub stars

ユースケース

Refining custom Claude Code agents for production-grade reliability and task success
Auditing legacy prompt files to migrate between Claude Code, OpenCode, and Codex
Securing AI agents by automatically identifying and restricting broad Bash tool access