Aegis

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

★ 1.3K67 forksPythonMITGanyuanRan

Aegis is a host-agnostic method pack of skills/plugins that makes AI coding agents plan against a project's real architecture baseline, verify completion with fresh evidence, and avoid unsafe or ghost-code changes. Ships installers for Codex, Claude Code, OpenCode, Cursor and many other skill-aware hosts.

Use Cases

Force AI coding agents to align with a project's real architecture baseline before editingRequire fresh verification evidence before accepting an agent's 'done' claimReduce rework and unsafe changes in long agentic coding tasksRoute tasks to strict/light/skipped TDD based on riskTrack and retire dead fallback code paths to stop silent tech debtKeep trivial tasks on a fast path while adding ceremony only when neededInstall one skill/method pack across Codex, Claude Code, OpenCode and other hostsRun an A/B benchmark comparing agent behavior with and without the method packBootstrap and verify skill discovery via a doctor script on a new hostEnforce first-principles, evidence-driven planning for coding agents

Built With

Language: Python
Frameworks: Claude Code · Codex · OpenCode · Cursor · Windsurf · GitHub Copilot · Kimi Code CLI · DeepSeek Harness · Trae · Qoder · ZCode · Grok Build · Pi CLI · OMP · OpenClaw · Hermes Agent

Aegis

Use Cases

Built With

Tags