Agent Blueprints

Each blueprint is a self-contained agent — system prompt, tool definitions, initialization code, and a walkthrough of how it handles a real task. Designed to work with OpenAI and Anthropic APIs. Framework-agnostic.

Note:

Start with the Research Agent. It uses the simplest tool set and demonstrates the ReAct loop clearly. Then move to Code Review (file system tools) and Content Writer (multi-step pipeline).

Prerequisites

All blueprints expect Python 3.10+ and an API key (OpenAI or Anthropic). Install the client library:

pip install openai anthropic

Research Agent

Web search + summarize + cite sources + fact-check loop. Tools: web_search, extract_content, fact_check.

Code Review Agent

Reads file tree, runs linter, checks patterns, suggests refactors. Tools: read_file, run_linter, list_directory, git_diff.

Content Writer

Outline → research → draft → edit → final. Tools: web_search, grammar_check, tone_adjust, seo_analyze.

Prompt Injection as Role Confusion — Why Style Trumps Tags

Why the ICML 2026 'role confusion' paper changes everything about how we defend LLM applications. The core insight: models identify text roles by writing style, not structural tags.

Permission-Gated Tool-Use: How Datasette Agent 0.3a0 Handles User Approval for Write Operations

A deep-dive tutorial on datasette-agent 0.3a0's execute_write_sql tool as a case study in safe agent tool design — covering the ask_user() approval pattern, permission gating, and how to build your own gated tools using the register_agent_tools hook.

CrewAI 3.0: Long-Term Memory, Tool Delegation, and RAG-Based Tool Selection

A hands-on tutorial for CrewAI 3.0's three flagship features — persistent long-term memory that survives across sessions, agent-to-agent tool delegation, and RAG-based dynamic tool discovery. Build a research crew that remembers past sessions and delegates tool calls between agents.

Agent Blueprints