CATEGORY · 65 ARTICLES

AI Coding Assistants

Working notes on the agents themselves — Claude Code above all, alongside Codex CLI, Copilot CLI, Cursor, Cline and Aider. The recurring subjects are the ones that decide whether a session goes well: what CLAUDE.md and AGENTS.md actually load, how a context window fills and what /compact and /clear really remove, MCP server configuration, hooks, skills, subagents, plan mode and permissions. Most of it is measured against real sessions rather than restated from documentation.

Five Claude Code extension mechanisms, with only hooks marked as a guaranteed action

FEATURED

AI Coding Assistants17 min read

Claude Code: The Complete Guide

Installation, CLAUDE.md, plan mode, hooks, MCP, and subagents — the mechanics of Claude Code, drawn from running it in production on a real repository.

Technical diagram illustrating Claude Code context window exhaustion and auto compaction thresholds

Claude Code Context Window Full: Diagnosis and Compaction Mechanics

Claude Code context window full warnings trigger at 160,000 tokens. We benchmarked the automatic compaction cycle, cache eviction economics, and zero-data-loss recovery workflows.

Oct 611 min read

AI Coding Assistants

Antigravity CLI installation diagram showing binary path and ~/.gemini config directory

How to Install Antigravity CLI: Architecture, Paths, and Setup

Antigravity CLI ships as a standalone Go binary rather than an npm package. Here is how to configure your PATH, manage ~/.gemini settings, and verify the install.

Oct 510 min read

AI Coding Assistants

Architecture diagram showing GitHub MCP server configuration and token overhead

MCP Servers GitHub Integration: Setup, Authentication, and Tool Overhead

Connecting the GitHub MCP server adds 26 tools to your agent. We measured the 4,820-token context tax, tested PAT scopes, and benchmarked it against gh CLI.

Oct 511 min read

AI Coding Assistants

Architectural comparison diagram between Model Context Protocol stdio pipes and REST HTTP APIs

MCP vs API: When to Use Model Context Protocol

When to build an MCP server vs calling a REST API directly from an AI agent. Measured payload sizes, stdio vs HTTP latency, and architectural trade-offs.

Oct 411 min read

AI Coding Assistants