Perdolique/workflow: A Complete Solution for Building Personalized AI Coding Workflows

Perdolique/workflow is an open-source collection of AI coding agents and skills designed to help developers build personalized AI coding workflows, addressing the pain point where general-purpose LLM assistants fail to meet specific project requirements. The project provides practical tools like Commit Agent, PR Agent, and TypeScript code style guidelines, supports deep integration with GitHub Copilot, and offers high customizability to help developers integrate LLM capabilities with real-world development practices.

With the widespread application of LLMs in software development, developers have found that general-purpose AI assistants struggle to meet the coding standards, tech stack preferences, and workflow requirements of specific projects. The Perdolique/workflow project was born out of this need, maintained by developer Perdolique. It brings together custom AI coding agents and reusable skills, focusing on practicality and customizability, allowing developers to create their own intelligent workflows.

The project architecture consists of two core components:

Custom AI Coding Agents

• Commit Agent: Analyzes code changes, generates commit messages that comply with conventional commit standards, and completes Git commits, enhancing the standardization of commit history.
• PR Agent: Analyzes branch differences, generates/updates GitHub PR content, simplifying the code review process.

Reusable Skills Collection

Includes 7 skills: code-style-typescript, commit-creator, drizzle-orm, markdownlint, playwright-e2e-testing, pr-creator, and vitest-unit-testing, providing professional knowledge and capability support.

The project offers multiple usage methods:

1. Installation via Vercel Skills CLI: pnpx skills add Perdolique/workflow to install all skills locally.
2. Agent Invocation: Invoke via AI coding assistants; the specific method varies by platform.
3. Deep Integration with GitHub Copilot: Override default behaviors using custom instruction files in the non-standard/copilot/instructions/ directory. It can be installed globally or at the project level (symbolic links required).

The project provides LSP configuration to support the /lsp command of GitHub Copilot CLI, with the configuration file path at non-standard/copilot/lsp-config.json. Currently, it supports TypeScript/JavaScript (using typescript-language-server) and Vue (using @vue/language-server) language servers. Installation methods are npm install -g typescript-language-server typescript and npm install -g @vue/language-server. Ensure the language servers are in the system PATH.

The project is positioned as a "personalized" toolbox, emphasizing practicality first, high customizability, and continuous evolution. Usage recommendations:

• Gradual adoption: Try a single skill first and observe its integration with existing workflows.
• Selectively enable agent functions and adjust according to team needs.
• Review Copilot instruction files to adapt to team standards.
• Contribute improvements or fork a customized version.

The project uses the Unlicense license (except for individual components), allowing free use, modification, and distribution. Although it is a personal project, its modular design gives it the potential to be a community resource. Developers can draw inspiration to build workflows suitable for their teams.

Perdolique/workflow represents a new paradigm in AI-assisted programming—a composable system composed of specialized components, better adapted to the complexity of modern software development. It is not just a toolset but also a way of thinking that demonstrates how to make AI adapt to developers' working styles. For teams exploring the integration of AI into their development processes, it is a good starting point and provides a reference for the future development of AI-assisted programming.