Rashedul Agent Engineering Workbench: Practices for Claude Code Skills, Sub-agents, and CI Workflow Reuse

This article introduces the Rashedul Agent Engineering Workbench, a personal template for agent engineering developers. It integrates Claude Code skill definitions, sub-agent orchestration patterns, and reusable CI/CD workflows, providing standardized practice references for AI-assisted software development and helping developers transform their collaboration with AI into a reusable and scalable engineering system.

With the popularity of AI coding assistants like Claude Code and GitHub Copilot, the software development paradigm has undergone changes, requiring developers to collaborate with AI. This new approach is called 'agent engineering', which emphasizes intent-driven (focus on expressing requirements), iterative collaboration (refining solutions through multi-round dialogues), and knowledge preservation (solidifying effective interaction patterns). This project is a summary of the author's practices in this field, organizing reusable skills, sub-agent configurations, and CI workflow templates.

Claude Code supports defining custom skills through the .claude/skills directory. Each skill is a structured instruction template (YAML format: name, description, prompt). Core skills include:

• Code review: covering security, performance, maintainability, and test coverage
• Refactoring assistant: analyzing code smells, formulating plans, executing refactoring, and verifying results
• Documentation generation: extracting APIs, generating specifications, writing examples, and updating documents
• Debugging assistant: collecting errors, analyzing paths, proposing hypotheses, and suggesting fixes

Complex tasks require collaboration among multiple agents. The project defines roles and processes:

• Roles: Architect (technical selection/architecture design), Developer (code implementation/testing), Reviewer (quality control), DevOps (deployment/monitoring)
• Collaboration process: Task decomposition → Parallel execution → Integration review → Iterative optimization → Delivery and deployment
• Context management: Shared workspace, structured communication (JSON format), memory persistence

GitHub Actions-based workflow templates:

1. AI-assisted code review: Triggered by PR, Claude reviews code and posts comments
2. Intelligent test generation: Scans uncovered code, AI generates test cases and submits PR
3. Document synchronization: When code changes affect APIs, automatically updates documents and deploys (Attached with YAML example structure for each workflow)

Key practices:

• Prompt engineering: Clear roles, provide context, specify output formats, iterative refinement
• Codebase organization: Standardized skill directories, templated prompts, example-driven
• Human-AI collaboration: AI draft + human refinement, divide complex tasks, human review for key decisions

Applicable scenarios: Individual developers improving efficiency, small teams standardizing collaboration, open source projects lowering entry barriers, education and training practice cases. Summary: This project transforms scattered AI-assisted behaviors into a reusable engineering system. It is a cutting-edge practice in AI-assisted software development and provides reference value for developers exploring AI coding assistants.