Agentic Coding

The @barlevalon/skills package provides a monorepo of reusable SKILL.md workflows specifically designed for AI-assisted engineering with agents like Claude. It offers portable skills that can be integrated into tools like Cursor, Continue, Cline, and VS Code Copilot through configuration files such as AGENTS.md rules or .cursor/rules. The collection includes workflows for PRD writing and planning (write-a-prd, prd-to-plan), debugging (diagnose), test-first development (tdd), code quality review, architecture improvement, and release preparation. Users can access skills via GitHub or install the npm package (@barlevalon/skills version 0.6.0), then invoke them by matching task descriptions to trigger relevant agent workflows.

Trellis is an engineering framework that addresses a core challenge in agentic coding: AI agents starting each session without project context or memory. It persists specifications, tasks, and project memory into your repository through a structured directory system (.trellis/spec/, .trellis/tasks/, .trellis/workspace/), allowing any coding agent to work consistently to your engineering standards. The framework implements a four-phase workflow—Plan (with PRD generation), Implement, Verify (auto-checking against specs and tests), and Finish (promoting learnings back to specs)—and works across 16 AI coding platforms including Claude Code and Cursor. For solo founders and small teams, Trellis enables repeatable "vibe coding" by writing conventions once and auto-injecting relevant context into each session, eliminating the need to re-specify requirements with every agent interaction and allowing team-shared standards to be versioned in the repo.

Gentle-AI is an ecosystem configurator that enhances AI coding agents like Claude Code and Cursor with persistent memory, Spec-Driven Development (SDD) workflows, curated coding skills, MCP servers, and per-phase model assignment. Rather than replacing agents, it supercharges existing ones by adding memory via Engram, allowing different AI models for different SDD phases (design, implementation, exploration), and providing structured delegation triggers to prevent monolithic sessions and maintain code quality through fresh reviews at key points like multi-file changes or commits. The project supports 15 agents and includes features like skill registry management, workspace-scoped configuration, and optional community plugins for OpenCode that provide real-time sub-agent activity monitoring and memory browsing directly from the IDE.