MG-Cafe

Security-hardened agentic coding assistant with 3 defense layers — input scanning, tool boundaries, and output scanning via Model Armor

Universal best-practice rules for Claude Code. Locks framework choices, layering boundaries, file locations, workflow patterns, and communication style. Drop into ./CLAUDE.md, .claude/skills/claude-code-best-practices/SKILL.md, or ~/.claude/skills/ for global use.

Budget-aware Claude Code to Codex handoff. Use this whenever the user mentions Claude Code limits, usage limits, rate limits, Agent SDK credits, claude -p automation budget, Codex rescue, /codex:rescue, handoff folders, or continuing work in Codex before Claude hits a limit.