wtf is agents?

This repo gives you a pack of 50 pre-configured AI assistants, each specialized for a different job, that plug into Claude Code (Anthropic's command-line coding tool). Instead of asking Claude a generic question and getting a generic answer, you can hand the task to a purpose-built expert, a security auditor, a database optimizer, a Python specialist, a content marketer, or even a legal advisor. The goal is to make Claude Code more useful by giving it specific roles rather than treating it as one generalist. Each subagent is a markdown file with a system prompt that defines what it knows and how it behaves. When you install them in the right folder, Claude Code can either pick the right specialist automatically based on what you're asking, or you can call one by name, like "use the code-reviewer to check my changes." The agents also coordinate with each other: ask it to build a feature and it might chain through an architect, then a frontend developer, then a tester, then a security auditor, all in sequence. The audience is developers and technical founders already using Claude Code who want deeper expertise without writing their own prompts. Practical examples include having a security auditor scan for vulnerabilities before a launch, a devops troubleshooter dig through production logs during an outage, or a content marketer draft a blog post about a new feature. What stands out is the model assignment strategy. Each agent is tagged to use a specific tier of Claude: the fast and cheap model for simpler tasks like documentation or social media posts, the mid-tier model for most coding work, and the most powerful model for hard problems like security audits and architecture review. This balances cost against capability so you're not paying for heavy compute on lightweight tasks. The README also references a separate companion project called Commands that pairs with these agents to trigger multi-step workflows via slash commands, though that's a different repo.