Stop rebuilding broken agent loops from scratch

A searchable pattern library of battle-tested Claude workflow patterns, each with a problem statement, flow diagram, copy-paste snippet, and documented failure modes.

Patterns that survive production

Every card covers a real failure mode found in production, not theory. Know what breaks before it breaks on you.

Search in under a minute

Client-side FuseJS search across all pattern cards. No account, no backend, no waiting.

Copy-paste ready snippets

Each pattern includes a working code snippet you can drop into your pipeline immediately.

FAQ

What patterns are covered?

Refine-Plan-Act loops, tool-calling patterns, context window management, error recovery, and parallelism. New cards added as patterns are validated in production.

Is this kept up to date?

Yes. Patterns are versioned in Markdown source on GitHub. You can submit corrections or new patterns via pull request.

Does it cover non-Claude LLM pipelines?

The patterns are Claude-focused but most apply to any tool-calling agent framework. Claude-specific quirks are called out explicitly.