GanyuanRan/Aegis

Make AI coding agents architecture-aware: baseline-first, evidence-verified, drift-checked, and safe across long tasks.

What it solves

Aegis addresses the tendency of AI coding agents to start writing code before they have a clear understanding of the goal, the project's architecture boundaries, the owner's requirements, or a reliable way to verify the changes. It provides a disciplined, evidence-driven workflow to prevent risky changes and ensure that simple tasks remain efficient without unnecessary overhead.

How it works

Aegis is a "method pack"—a set of composable skills and workflow guardrails that can be installed globally across various AI coding hosts (such as Claude Code, GitHub Copilot, or Trae). It implements a baseline-first approach where planning and evidence gathering must precede completion claims. It includes specialized modes like "Grill me" for decision interviews, strict TDD (Test-Driven Development) routing, and first-principles reviews to pressure-test complex directions before implementation.

Who it’s for

Software engineering teams and individual developers who use AI coding agents for real-world software projects and want to move beyond basic code generation toward a more reliable, architecture-aware development process.

Highlights

  • Multi-Host Support: Compatible with a wide range of AI coding agents and CLIs including Claude Code, DeepSeek-TUI, and GitHub Copilot.
  • Evidence-Driven Workflow: Requires evidence of success before a task is marked as complete.
  • Decision Guardrails: Features a "Grill me" mode to force one-at-a-time decision questioning to avoid premature implementation.
  • Flexible TDD: Supports automatic or strict TDD routing based on task risk.
  • Architecture Focus: Encourages baseline-first planning and the use of domain modeling to resolve ambiguous terms or conflicting project rules.