Claude Cookbook

Anthropic’s Claude Cookbook is on Hacker News with a set of guides for building on Claude: programmatic tool calling (PTC), tool search with embeddings, automatic context compaction, a crop tool for image analysis, prompting for frontend aesthetics, and an introduction to Claude Skills. A related cookbook entry covers reproducing Claude’s agentic search benchmark scores in the Messages API (Jun 2026).

On the technical side, PTC has Claude write code that calls tools inside a code execution environment so work happens programmatically rather than through repeated back-and-forth tool turns, which is aimed at cutting latency and token use. Tool search with embeddings uses semantic embeddings so the model can discover the right tools dynamically when an app exposes thousands of them. Automatic context compaction compresses conversation history so long-running agentic workflows stay inside context limits. The crop tool lets Claude zoom into regions of charts, documents, and diagrams for finer-grained image analysis.

For engineers and builders, the cookbook is less a single product drop and more a pattern library for production agent stacks: tool-heavy apps, long sessions, multimodal document work, and frontend generation. PTC and embedding-based tool discovery address two common failure modes—tool-call chatter and flat tool lists that do not scale—while context compaction targets the hard limit of stuffing full history into every step.

In market terms, this sits in the same race as other major model platforms that are shipping agent runtimes, tool ecosystems, and skills-style document automation. Claude Skills (Excel, PowerPoint, PDF) compete on office-workflow automation; the frontend aesthetics guide and crop tool push on design quality and visual document understanding, where many generic agent demos still look weak.

Practical takeaway: if you already wire Claude through the Messages API, treat PTC, tool search, and compaction as first-class design choices—not optional cookbook demos—especially when tool count, session length, or image-heavy inputs are the bottleneck. Watch how teams combine embedding-based tool discovery with PTC in real multi-tool agents, and whether Skills become the default path for spreadsheet, deck, and PDF automation rather than one-off prompts.