How carlosnavarrovalldecabres-vlc/agents works
Project Type and Purpose
• This is an open-source “plugin” or agent-definition library for Claude Code. Each Markdown file describes an AI agent (name, description, tools, system prompt). The repository’s purpose is to organize, document, and distribute these agents so that Claude Code can load and invoke them automatically or on demand.
File Structure Analysis
• Top-level:
• .gitignore, README.md – repo metadata
• design/, engineering/, marketing/, product/, project-management/, studio-operations/, testing/, bonus/ – department folders
• Each folder contains .md files with YAML frontmatter.
• Installation involves a simple copy/clone operation; there is no build script or compiled code.
• The “engine” that consumes these files resides in Claude Code; this repo is purely content.
Key Insights from README
• Agents are categorized by department for discovery.
• Installation: clone & copy into ~/.claude/agents/; restart Claude Code.
• Agents auto-register and trigger proactively based on context.
Essential Architectural Elements to Capture
• Separation of concerns: content repo vs. runtime vs. installation vs. UI.
• Plugin loading process (file → parse → register).
• Agent dispatch logic (intent matching, proactive triggers).
• Extensibility workflow (adding/modifying .md files).
• Performance metrics pipeline (runtime → logging → analytics).