What it does

LazyCodex installs OmO (oh-my-openagent) as a set of Codex commands. You get $ulw-plan for strategic planning, $start-work for checklist-driven execution, and $ulw-loop for self-referential tasks that run until an Oracle agent verifies completion. It also adds /init-deep to generate hierarchical AGENTS.md files that give future agents landmarks in large codebases.

The interesting bit

The model routing is the quietly clever part. OmO doesn’t burn your best model on every subtask—it categorizes work (quick, ultrabrain, agentic coding) and routes to appropriate GPT models with fallback chains. The README cites specific OpenAI model docs for each tier, which is more rigor than most agent projects bother with.

Key highlights

• One-line install via npx lazycodex-ai install; no global npm packages
• Project memory through hierarchical AGENTS.md generation (/init-deep)
• Verified completion loops capped at 500 iterations (ultrawork) or 100 (normal)
• Extensible skill system: LSP diagnostics, AST-grep, comment-checker, UI/UX polish, “remove-ai-slops” cleanup
• Multi-model routing with quota discipline—cheaper models for small edits, frontier models for hard logic
• Thin wrapper: core engine is OmO submodule, plus a Next.js 15 web package for docs

Caveats

• The README is upfront that this is “a thin distribution layer”—if you want the core engine, that’s OmO itself
• Some command documentation lives at an external site (lazycodex.ai/docs) rather than in-repo
• The “60K Stars” claim refers to OmO’s social media traction, not LazyCodex’s 547 GitHub stars

Verdict

Worth a look if you’re already using Codex and want structured planning, execution, and verification without building your own agent harness. Skip if you want a standalone tool—this is packaging and configuration, not a from-scratch system.