Skip to main content

A LIGHT PORT OF OMO'S HEPHAESTUS, FOR CODEX

LazyCodex

The Hephaestus deep-worker agent, ported light into Codex.

A focused slice of OmO — goals not recipes, verified completion.

  • goals not recipes
  • parallel exploration
  • verified completion

Ultrawork, live

Watch an ultrawork run close the loop

Include ultrawork (or the short alias ulw) anywhere in your prompt and the harness switches to maximum-precision, outcome-first, evidence-driven orchestration. An agent saying it is done does not mean the work is done — the work is done when observable evidence verifies it.

ulw add authentication
Plan, execute, verify, and keep the evidence attached.
01 Research
ulw add authentication

ULTRAWORK MODE ENABLED!

Research wave · Explorer and Librarian gather source truth
task(subagent_type="explorer") + task(subagent_type="librarian")

Ultrawork starts by fanning out research.

Root does not guess. Explorer reads the local implementation, Librarian checks docs and contracts, and their findings become the input to the Plan agent.

npx lazycodex-ai install

= npx --yes --package oh-my-openagent omo install --platform=codex

$ulw-loop

$ulw-loop "task" [--strategy=reset|continue]

A self-referential loop that runs until VERIFIED completion.

  • Emits <promise>DONE</promise> when it reaches a completion claim
  • Oracle must verify before the loop ends
  • Iteration cap 500 in ultrawork mode (100 normal)

$ulw-plan

$ulw-plan "what to build"

Prometheus strategic planner — turns an idea into a decision-complete work plan.

  • Socratic interview, then parallel codebase exploration
  • Metis gap analysis, optional Momus review
  • Writes plans/<slug>.md — never writes product code

$start-work

$start-work [plan] [--worktree <path>]

Executes a Prometheus plan until every checkbox is done.

  • Durable Boulder state survives across turns
  • Parallel subagents, strict TDD + 5 evidence gates
  • Prints ORCHESTRATION COMPLETE when finished

What LazyCodex wires into Codex

Harness the whole codebase

LazyCodex installs OmO as a serious agent harness for complex repositories: project memory, planning, execution, skills, hooks, model routing, and verification defaults in one pass.

01

Context that survives

$init-deep generates hierarchical AGENTS.md context so agents start from local guidance before touching a large repository.

02

Plans before edits

$ulw-plan turns ambiguous work into a decision-complete plan, then $start-work executes it with durable progress tracking.

03

Evidence at the end

$ulw-loop keeps complex work moving until the completion promise is backed by verification, not a hopeful status update.

Built-in skill coverage

Skills auto-activate when a task matches their domain, so you do not need to study every one first. Add a skill name to your prompt when you want to call it explicitly; ulw-research is the maximum-saturation mode for deep codebase, web, official-docs, and OSS-repo research.

  • ulw-research
  • review-work
  • remove-ai-slops
  • frontend
  • programming
  • visual-qa
  • LSP
  • AST-grep

Team Mode

Run a named team of cooperating Codex threads

One leader, durable state on disk. The main session is always the team leader: it splits the work and assigns each slice, holds live situational awareness of every member, verifies and QAs what they deliver, relays findings between members, and synthesizes the result.

Members are defined by a concrete part, ownership area, or perspective — never a vague job role.

A bundled cross-platform script writes the .omo/teams state plus an auto-generated member field manual.

Triage feature and question issues
Review PR readiness
Triage LazyCodex issues
Triage runtime bug reports
Triage feature and question issuesrunning
Review PR readinessrunning
Triage LazyCodex issuesreported
Triage runtime bug reportsreported

Sent by Codex from another thread

Member A COMPLETE verification note: report exists, pinned-link check passed, and no GitHub mutations/repo edits.

When a team beats plain subagents

  • The work does not split into perfectly isolated pieces, but doing it in parallel is clearly more convenient — members need to see and react to each other's findings.
  • One task still needs exploration, yet its goal is already clear — parallel investigation under a fixed objective.

$ulw-research

Maximum-saturation research orchestration

Parallel explore and librarian swarms across the codebase, web, official docs, and OSS repos; a recursive expand loop driven by the leads workers return; empirical verification by running code; cited synthesis and optional reports.

  • codebase
  • web
  • official docs
  • OSS repos

Activates only on an explicit demand for research — say ulw-research or any ulw research wording in your prompt. While active, exhaustive coverage is the goal.

Where it comes from

A light port of OmO's Hephaestus

OmO is the full agent harness — a primary orchestrator, a deep worker, specialist agents, multi-model routing, and dozens of lifecycle hooks. LazyCodex does not ship all of that. It ports one role, Hephaestus, into Codex as a focused, repeatable setup: the deep worker plus the workflows that keep its runs honest.

OmO — the full harness

  • Sisyphus orchestrator with boulder.json session continuity
  • Hephaestus deep worker + specialist agents
  • 54+ lifecycle hooks, multi-model routing, team mode

LazyCodex — the Hephaestus port

  • Hephaestus only: goals, not step-by-step recipes
  • $ulw-plan / $start-work / $ulw-loop workflows
  • Skills, hooks, and verification defaults for Codex
Deep Worker

Hephaestus

Give it a goal, not a recipe

Named after the Greek god of the forge. Methodical, thorough, obsessive. Fires parallel explore subagents before writing any code, then runs a tight loop until the work is proven. Built for deep architectural reasoning, complex debugging, and cross-domain synthesis.

  1. 01

    Explore

    Map the terrain — read with tools, never speculate.

  2. 02

    Plan

    Chart the course — files to change, specific edits, dependencies.

  3. 03

    Implement

    Build with precision — surgical edits that match codebase style.

  4. 04

    Verify

    Prove it works — LSP diagnostics, tests, build, in parallel.

  5. 05

    Manually QA

    Drive the real surface — HTTP, tmux, browser — then report.

For when good enough isn't.

Ultrawork

Ultrawork

Ready to wire the harness?

Install the Codex setup, then run the workflows that keep large changes under control.