Agentic Coding process This repository contains my current process for working with agentic coding assistants. It's focused around spec driven development and has a number of helpful tools and commands. Feel free to use what you see here and/or propose improvements. The framework now includes cross platform support with GitHub Copilot prompt files, forge aware review tooling that runs on both GitHub and GitLab, peer review that adapts to whether you're on a personal or a work machine, and improved workflow management with the rune CLI integration. Agents The framework provides specialized AI agents for different aspects of development. These are defined for use in Claude Code, and not every tool supports sub agents like this. Some agents lean on external AI systems when available — peer review validator consults the Codex/Kiro MCP agents, but only on personal machines (see Personal vs. work projects); elsewhere it falls back to Claude subagents so no code leaves the machine. code simplifier Reviews code for complexity reduction and maintainability improvements design critic Provides critical review of design documents and architecture proposals (now using Sonnet model for improved efficiency) efficiency optimizer Analyzes code for performance optimization opportunities local review Sonnet powered local replacement for an automated CI review step. Forge aware: it detects GitHub or GitLab from the git remote and reviews the open change request (code quality, bugs, performance, security, test coverage, plus whatever the project's CLAUDE.md mandates), posting a single review comment instead of consuming private CI/Actions minutes. See Forge support peer review validator Validates decisions by obtaining at least two independent peer perspectives. On personal machines ( PERSONAL PROJECTS=1 ) it consults external AI systems (Codex, Kiro); otherwise it spawns Claude subagents with distinct lenses so the same balanced review runs without sending code off machine. See Persona