Give coding agents the expertise of specific roles with 411 skills
This collection of instructions is designed to make AI coding agents follow the work methods of professionals, from software engineers to CTOs and regulatory experts. The core idea is to make agents behave like domain experts in response to plain natural language questions. The repository provides 411 verified skills and 807 Python scripts that are ready to use in more than 13 coding agents, including Claude Code, Codex, and Gemini CLI.
Role-specific installation
Users select and install skill bundles that match their role. Software engineers can install engineering-skills, which focuses on architecture and code review, while product managers can install product-skills, and those in regulated industries can install ra-qm-skills, which contains knowledge related to ISO 13485 or the FDA. Each bundle includes the workflows and decision-making frameworks that the role typically handles. In Claude Code, you can apply them simply with the /plugin install command, while other agents use the npx skills add command.
Differences between skills, agents, and personas
The repository distinguishes and provides three concepts. A skill is a modular instruction package that defines how to perform a specific task. An agent is the entity that performs tasks in a specific domain, while a persona defines the identity that determines how the agent thinks and prioritizes. For example, the 'Senior Architect' skill provides architecture review procedures, whereas the 'Startup CTO' persona applies a strategic perspective when selecting a technology stack. These three elements are combined through an orchestration protocol to handle complex projects.
Verified skills and security audits
All skills are tested to ensure they load correctly when requested in plain language. Specifically, skill-security-auditor acts as a security gate that scans for malicious code or prompt injection risks before skill installation. This tool checks for six risk factors, including command injection, data leakage, and privilege escalation, and returns PASS, WARN, or FAIL results. Additionally, agent-run-forensics verifies through session logs whether the work the agent actually performed matches the work it claimed to have done.
Support for conversion to various tools
A single script execution can convert skills into the formats of more than nine other AI coding tools, such as Cursor, Aider, and Windsurf. The converted files are installed in each tool's rules directory, requiring no manual conversion work. The 807 scripts use only the Python 3 standard library, making them executable in any environment.
Pre-application checks
This repository focuses on porting the work methods of specific roles to AI agents. Because it covers fields where precision is critical, such as regulatory compliance and financial management, it is recommended to have a human review the skill outputs before applying them to actual work. Windows users should check symbolic link settings and UTF-8 encoding environment variables, and some skills are designed to run locally without external API calls. It is advisable to determine which of the repository's 20 categories your work domain falls into, then install and test the corresponding bundle.
