lazycodex

LazyCodex is a distribution layer that packages OmO (oh-my-openagent) as an agent harness for OpenAI's Codex CLI, aimed at complex codebases. It installs via `npx lazycodex-ai install` (no global npm install) and provides a set of skills, hooks, agent roles, and commands that extend Codex with project memory, planning, execution, and verified completion. Key features include: - **$init-deep**: Generates hierarchical AGENTS.md context for large repositories, scoring complex directories and writing local guidance. - **$ulw-plan**: A strategic planner that writes decision-complete plans to `plans/<slug>.md` without touching product code. - **$start-work**: Executes a plan checklist with durable progress tracking until all checkboxes are done. - **$ulw-loop**: A self-referential loop that runs until Oracle-verified completion, with iteration caps (500 in ultrawork mode, 100 in normal mode). - **Sub-agent roles**: Installs selectable roles (explorer, librarian, plan, momus, metis, codex-ultrawork-reviewer) into `~/.codex/agents/`, usable via Codex's `spawn_agent` tool with `agent_type`. - **Multi-model routing**: Automatically selects models per task category (e.g., quick tasks to gpt-5.4-mini, hard logic to high-reasoning models) to manage quota. - **Skills system**: Includes review-work, remove-ai-slops, frontend-ui-ux, programming (TypeScript/Rust/Python/Go), LSP, AST-grep, rules, and comment-checker. - **Hooks & lifecycle**: Pre/post hooks for agent actions, with approval required before running. Installation options include a no-TUI autonomous mode (`--no-tui --codex-autonomous`) and an experimental Codex marketplace path. A `doctor` command checks installation health, and `uninstall` removes all managed components. The project is MIT-licensed and maintained by Sisyphus Labs, with the core engine OmO included as a submodule.