A curated collection of resources, patterns, and templates for building reliable scaffolding for agents. Harness engineering is the discipline of designing the systems surrounding an agent—such as context delivery, tool interfaces, planning artifacts, verification loops, memory systems, and sandboxes—that determine its success or failure on real tasks. This focus is on the harness rather than the model.
- Design primitives for loops, planning, and memory
- Reference implementations and tutorials
- Security, sandboxing, and permissions
- Evaluation, verification, and observability
- Task runners and orchestration
- Human-in-the-loop and production operations
- Design primitives for loops, planning, and memory
- Reference implementations and tutorials
- Security, sandboxing, and permissions
- Evaluation, verification, and observability
- Task runners and orchestration
- Human-in-the-loop and production operations
This tutorial provides a step-by-step guide to building a lightweight personal AI agent inspired by the nanobot architecture in Google Colab. The approach focuses on recreating core components—such as provider abstractions, tool registration, session memory, and lifecycle hooks—rather than relying on heavy external frameworks. Key features include a tool registry for Python functions, token-budgeted memory management, and an MCP-style tool server for external capabilities. The guide includes a complete Python implementation that supports both live OpenAI-compatible models and a deterministic mock provider for offline testing.
Main topics covered:
- Provider abstraction for multi-model compatibility
- Automated tool schema generation using decorators
- Session-specific memory with token budgeting
- Lifecycle hooks for auditing and timing
- Dynamic skill loading and MCP server connection
Main topics covered:
- Provider abstraction for multi-model compatibility
- Automated tool schema generation using decorators
- Session-specific memory with token budgeting
- Lifecycle hooks for auditing and timing
- Dynamic skill loading and MCP server connection
llayer applies the Unix philosophy to large language model orchestration by building framework-free agents with bash, curl, and jq. The architecture decomposes the agent lifecycle into three fundamentals: an append-only JSONL history file for state and memory, a jq stream reducer for context window management, and a standard bash while loop for control flow. This stateless text pipeline enables time-travel debugging via simple file slicing, zero abstraction tooling through native bash functions, and seamless POSIX tool integration for filtering or benchmarking. The system functions as a REPL-style loop that ingests user input, constructs context, evaluates it against a local model like Ollama, handles tool dispatches, and outputs results. All interactions are recorded immutably in a structured JSONL event schema, prioritizing transparency, composability, and minimalist design.
- Append-only JSONL history for auditing and replayability
- Modular command chaining for stateless and stateful interactions
- Docker Compose integration for local Ollama inference
- Transparent POSIX tool pipeline for data filtering and token benchmarking
- Minimalist schema with explicit event types and sources
- Append-only JSONL history for auditing and replayability
- Modular command chaining for stateless and stateful interactions
- Docker Compose integration for local Ollama inference
- Transparent POSIX tool pipeline for data filtering and token benchmarking
- Minimalist schema with explicit event types and sources
An examination of the hype surrounding autonomous AI agent frameworks and why they may add unnecessary complexity to software development. The author argues that for most production use cases, structured workflows using LLM function calling are more reliable than fully autonomous agents.
- Complexity vs control in agentic systems
- Limitations of current models regarding long-term autonomy
- Advantages of explicit programming over unpredictable loops
- Complexity vs control in agentic systems
- Limitations of current models regarding long-term autonomy
- Advantages of explicit programming over unpredictable loops
Google DeepMind has released the Gemma 4 12B, a dense multimodal model featuring an encoder-free architecture. Unlike previous iterations that used separate vision and audio encoders, this model allows these modalities to flow directly into the LLM backbone. This streamlined design reduces latency and memory overhead, allowing the model to perform agentic reasoning tasks on consumer laptops with as little as 16 GB of VRAM while approaching the performance levels of much larger models like the 26B MoE variant.
- Unified decoder-only architecture for text, image, video, and native audio input.
- Encoder-free design using a 35M vision embedder and direct raw audio wave projection.
- Optimized to run locally on Apple Silicon Macs and consumer GPU laptops.
- Released under an Apache 2.0 license with support for llama.cpp, MLX, vLLM, and Ollama.
- Unified decoder-only architecture for text, image, video, and native audio input.
- Encoder-free design using a 35M vision embedder and direct raw audio wave projection.
- Optimized to run locally on Apple Silicon Macs and consumer GPU laptops.
- Released under an Apache 2.0 license with support for llama.cpp, MLX, vLLM, and Ollama.
Anthropic shares insights gained from developing and scaling hundreds of internal skills for Claude Code. The article defines skills as collections of instructions, scripts, and resources that help AI agents perform tasks more accurately and efficiently. It provides a framework consisting of nine distinct skill categories used within Anthropic and offers practical advice on designing effective skills, such as including gotchas sections and writing descriptions optimized for models rather than humans.
- Definition and structure of agentic skills
- Nine functional categories for skill organization
- Best practices for skill design and implementation
- Strategies for distributing and managing a skills marketplace
- Definition and structure of agentic skills
- Nine functional categories for skill organization
- Best practices for skill design and implementation
- Strategies for distributing and managing a skills marketplace
Open Code Review is an AI-powered CLI tool designed for automated, high-precision code reviews. Originally developed as Alibaba Group's internal assistant, the project uses a hybrid architecture that combines deterministic engineering with LLM agents to provide stable and accurate feedback. Unlike general-purpose agents, it employs smart file bundling and fine-grained rule matching to maintain context and prevent issues like position drift or incomplete coverage on large changesets.
Key features:
- AI-driven line-level review comments
- Hybrid architecture combining hard constraints with dynamic decision-making
- Support for various LLM endpoints including OpenAI and Anthropic
- Seamless integration with CI/CD pipelines and coding agents like Claude Code
- Customizable rule sets for specific project requirements
Key features:
- AI-driven line-level review comments
- Hybrid architecture combining hard constraints with dynamic decision-making
- Support for various LLM endpoints including OpenAI and Anthropic
- Seamless integration with CI/CD pipelines and coding agents like Claude Code
- Customizable rule sets for specific project requirements
> Lessons from building a fast, reliable scientific agent with local open-weight models, vLLM, and long-context infrastructure
Google is transitioning from the Gemini CLI to the new Antigravity CLI, a core component of the Google Antigravity agent-first development platform. This shift addresses the growing need for multi-agent orchestration and unified backends in developer workflows. The new tool provides faster execution using Go and supports asynchronous background tasks for complex operations like large-scale refactoring or research.
Key points:
* Transitioning from Gemini CLI to Antigravity CLI
* Introduction of the Google Antigravity agent-first platform
* Faster, Go-based performance and asynchronous workflow support
* Sunset dates for consumer services starting June 18, 2026
* Continued support for enterprise customers through existing licenses
Key points:
* Transitioning from Gemini CLI to Antigravity CLI
* Introduction of the Google Antigravity agent-first platform
* Faster, Go-based performance and asynchronous workflow support
* Sunset dates for consumer services starting June 18, 2026
* Continued support for enterprise customers through existing licenses
This tutorial demonstrates how to evolve a standard chatbot into a truly agentic system using the Gemma 4 model family. Instead of relying solely on remote web APIs, it shows how to provide the model with tools that interact directly with the local environment—specifically a sandboxed filesystem explorer and a restricted Python interpreter. By implementing security measures like path-traversal guards for file access and whitelisted builtins for code execution, users can safely allow small models running locally on laptops to observe their surroundings and perform deterministic calculations.
Main topics:
* Transitioning from API retrieval to true agency through local system interaction.
* Building a secure filesystem explorer with path-traversal protection.
* Implementing a restricted Python interpreter using exec() and whitelisted builtins.
* Orchestrating tool calls using Gemma 4 and Ollama for local agentic workflows.
Main topics:
* Transitioning from API retrieval to true agency through local system interaction.
* Building a secure filesystem explorer with path-traversal protection.
* Implementing a restricted Python interpreter using exec() and whitelisted builtins.
* Orchestrating tool calls using Gemma 4 and Ollama for local agentic workflows.
