Skip to content
/

oh-my-codex (sigridjineth)

oh-my-codex-sigridjineth · sigridjineth/oh-my-codex · ★ 14 · last commit 2026-03-09

A personal Codex plugin extension adding tier-based model routing, LSP/AST tooling, and notification integrations to the canonical oh-my-codex workflow.

Best whenCost-aware model tier selection (Haiku for simple, Sonnet for standard, Opus for complex) should be explicit and deterministic, not heuristic.
Skip ifSerializing independent work, Routing all tasks to Opus regardless of complexity
vs seeds
superpowers(skills-first) and taskmaster-ai (multi-model routing). The unique contribution over any seed is the explicit Haiku/Sonn…
Primitive shape 73 total
Commands 6 Skills 41 Subagents 20 Hooks 6
00

Summary

oh-my-codex (sigridjineth) — Summary

oh-my-codex by sigridjineth (Sigrid Jin) is the personal Codex plugin of a named Ambassador of the canonical oh-my-codex project (Yeachan-Heo). With 14 stars and active maintenance (last commit March 2026), it is the most feature-rich of the three in terms of skills surface area (41 skills vs 46 canonical/14 scalarian) and adds unique components not present in the canonical project: LSP/AST tooling, Discord/Telegram notification integration, rate-limit recovery, project session manager, multi-tier model routing (Haiku/Sonnet/Opus based on task complexity), and a docs/shared/agent-tiers.md model-tier routing reference. Sigrid is listed as an Ambassador and contributor in the canonical OMX project, making this a derivative with real added value. Key additions: configure-discord, configure-telegram, tier-aware ultrawork (fires Task(subagent_type="oh-my-codex:executor-low", model="haiku") for simple changes), and 20+ agent personas. Closest seed comparison: resembles superpowers in skills-first design but adds the most sophisticated multi-model routing in this batch, explicitly using Haiku/Sonnet/Opus tiers based on task complexity.

01

Overview

oh-my-codex (sigridjineth) — Overview

Origin

Created by Sigrid Jin (@sigridjineth), who is named as an Ambassador in the canonical oh-my-codex project (Yeachan-Heo). This is described as "Sigrid's personal codex plugin." The npm package is published as oh-my-codex (same name as canonical — installable via Codex plugin marketplace from this repo). Last commit March 9, 2026.

Philosophy

From README:

"oh-my-codex is designed to make Codex behave like a practical engineering conductor: plan first, delegate by role, verify outputs, detect drift, and keep sessions resumable."

"Base Codex is excellent at direct implementation. oh-my-codex adds a full orchestration layer for larger or riskier workflows: Reusable specialized agents with explicit responsibilities, Multi-step planning and review-first patterns, Enforcement and verification hooks, Parallel execution modes for speed, Type-aware and syntax-aware tooling (LSP + AST), Runtime utilities for drift detection and resumable sessions."

Unique Additions vs Canonical

  1. Multi-tier model routing: docs/shared/agent-tiers.md defines LOW/MEDIUM/HIGH tier → Haiku/Sonnet/Opus mapping
  2. LSP tooling family: goto-definition, find-references, symbols, diagnostics, rename
  3. AST search/replace tools: structural code edits
  4. Discord/Telegram notifications: $configure-discord, $configure-telegram
  5. Rate-limit recovery: tmux-based wait/resume for Codex API rate limits
  6. Project session manager ($project-session-manager, $psm): drift detection, resumable sessions
  7. LSP + AST MCP integration: developer tooling beyond pure workflow skills
  8. HUD statusline support: integration with shell status lines

Execution Modes

Mode Purpose
autopilot Broad autonomous execution
ultrawork Maximum parallelism with tier routing
ultrapilot Parallel autopilot with ownership patterns
swarm N-agent coordinated execution
pipeline Ordered stage handoff
ecomode Cost-sensitive execution
ralph Single-owner persistence with verification
ultraqa QA-heavy validation cycles
02

Architecture

oh-my-codex (sigridjineth) — Architecture

Distribution

  • Codex plugin (preferred): codex plugin marketplace add https://github.com/sigridjineth/oh-my-codex
  • CLI install from source (fallback): git clone && npm install && npm run build && node dist/cli/index.js install
  • Installs into ~/.codex/ (hooks, agents, CODEX.md, HUD config)

Required Runtime

  • Node.js >= 20
  • Codex CLI

Directory Tree

oh-my-codex/
├── src/
│   ├── installer/     — installation, migration, conflict handling
│   ├── hooks/         — lifecycle hook logic
│   ├── features/      — execution modes and runtime features
│   ├── hud/           — usage telemetry and statusline
│   └── ...
├── agents/            — 20+ agent persona .md files
├── skills/            — 41 SKILL.md files
├── commands/          — command definitions
├── hooks/
│   └── hooks.json     — lifecycle hooks definition
├── scripts/           — setup/update/runtime scripts
│   ├── keyword-detector.mjs
│   ├── skill-injector.mjs
│   ├── session-start.mjs
│   └── project-memory-session.mjs
├── docs/
│   └── shared/
│       └── agent-tiers.md  — tier routing reference (LOW/MEDIUM/HIGH)
├── bridge/            — bridge/integration layer
├── research/          — research skill artifacts
└── dist/              — compiled output

Target AI Tools

  • Primary: OpenAI Codex CLI
  • Hooks designed for Codex plugin system (.codex-plugin/)

State Root

~/.codex/ for global install
Per-project: unspecified (sessions persist to .codex/ structure)

Key Files

  • hooks/hooks.json — defines UserPromptSubmit, SessionStart hooks with keyword detector and skill injector
  • docs/shared/agent-tiers.md — model tier routing reference
  • dist/cli/index.js — CLI entry point with install, doctor conflicts commands
03

Components

oh-my-codex (sigridjineth) — Components

CLI (minimal)

  • node dist/cli/index.js install — install to ~/.codex/
  • node dist/cli/index.js doctor conflicts --json — conflict check

No omx binary published; uses Codex CLI directly with plugin install.

Skills (41 total)

Skill Purpose
ultrawork Parallel execution engine with tier-based model routing
autopilot Broad autonomous execution
ultrapilot Parallel autopilot with ownership patterns
ultraqa QA-heavy validation cycles
swarm N-agent coordinated execution with claim/checkpoint
pipeline Ordered stage handoff
ecomode Cost-sensitive execution
ralph Single-owner persistence loop
ralph-init Ralph initialization
team Team coordination
plan Planning workflow
ralplan Plan approval
analyze Code analysis
deepsearch Deep search
deepinit Deep project initialization
research Research workflow
code-review Code review
security-review Security review
tdd TDD enforcement
trace Code tracing
design Design workflow
build-fix Build failure recovery
git-master Git workflow
hud HUD state
doctor Health check
omc-setup OMC setup
skill Skill authoring
writer-memory Writer memory for sessions
note Note-taking
help Help routing
cancel Cancel operations
frontend-ui-ux Frontend workflow
configure-discord Discord notification setup
configure-telegram Telegram notification setup
mcp-setup MCP server configuration
release Release workflow
learn-about-omc OMC onboarding
learner Learning mode
project-session-manager Drift detection, resumable sessions
psm Alias for project-session-manager
AGENTS.md Agents directory index

Agent Personas (agents/ directory, 20+ files)

analyst, api-reviewer, architect, build-fixer, code-reviewer, critic, debugger, deep-executor, dependency-expert, designer, executor, explore, git-master, information-architect, performance-reviewer, planner, product-analyst, product-manager, qa-tester, quality-reviewer, (+ more)

Hooks (hooks.json)

Event Handler Purpose
UserPromptSubmit keyword-detector.mjs Detect OMC keywords and route
UserPromptSubmit skill-injector.mjs Inject skill content
SessionStart (startup/clear/compact) session-start.mjs Initialize session state
SessionStart (startup/clear/compact) project-memory-session.mjs Load project memory
SessionStart (init) setup-init.mjs Project initialization
SessionStart (maintenance) (additional handlers) Maintenance mode

Developer Tooling (via skills/commands)

  • LSP family: definitions, references, symbols, diagnostics, rename
  • AST search/replace: structural code edits
  • Rate-limit recovery: tmux-based wait/resume
  • HUD statusline integration
05

Prompts

oh-my-codex (sigridjineth) — Prompts

Verbatim Excerpt 1: ultrawork/SKILL.md (Tier-Based Execution)

Prompting technique: Cost-aware model tier routing; parallel-first dispatch; background operation discipline

<Execution_Policy>
- Fire all independent agent calls simultaneously -- never serialize independent work
- Always pass the `model` parameter explicitly when delegating
- Read `docs/shared/agent-tiers.md` before first delegation for agent selection guidance
- Use `run_in_background: true` for operations over ~30 seconds (installs, builds, tests)
- Run quick commands (git status, file reads, simple checks) in the foreground
</Execution_Policy>

<Steps>
1. Read agent reference: Load `docs/shared/agent-tiers.md` for tier selection
2. Classify tasks by independence: Identify which tasks can run in parallel vs which have dependencies
3. Route to correct tiers:
   - Simple lookups/definitions: LOW tier (Haiku)
   - Standard implementation: MEDIUM tier (Sonnet)
   - Complex analysis/refactoring: HIGH tier (Opus)
4. Fire independent tasks simultaneously: Launch all parallel-safe tasks at once
5. Run dependent tasks sequentially: Wait for prerequisites before launching dependent work
6. Background long operations: Builds, installs, and test suites use `run_in_background: true`
7. Verify when all tasks complete (lightweight):
   - Build/typecheck passes
   - Affected tests pass
   - No new errors introduced
</Steps>

<Tool_Usage>
- Use `Task(subagent_type="oh-my-codex:executor-low", model="haiku", ...)` for simple changes
- Use `Task(subagent_type="oh-my-codex:executor", model="sonnet", ...)` for standard work
- Use `Task(subagent_type="oh-my-codex:executor-high", model="opus", ...)` for complex work
- Use `run_in_background: true` for package installs, builds, and test suites
</Tool_Usage>

Verbatim Excerpt 2: hooks.json (Runtime Hook Architecture)

Prompting technique: Keyword detection → skill injection pipeline; session memory priming on every start

{
  "description": "OMC orchestration hooks with async capabilities",
  "hooks": {
    "UserPromptSubmit": [
      {
        "matcher": "*",
        "hooks": [
          {"type": "command", "command": "node \"${CODEX_PLUGIN_ROOT}/scripts/keyword-detector.mjs\"", "timeout": 5},
          {"type": "command", "command": "node \"${CODEX_PLUGIN_ROOT}/scripts/skill-injector.mjs\"", "timeout": 3}
        ]
      }
    ],
    "SessionStart": [
      {
        "matcher": "*",
        "hooks": [
          {"type": "command", "command": "node \"${CODEX_PLUGIN_ROOT}/scripts/session-start.mjs\"", "timeout": 5},
          {"type": "command", "command": "node \"${CODEX_PLUGIN_ROOT}/scripts/project-memory-session.mjs\"", "timeout": 5}
        ]
      }
    ]
  }
}

Key Difference from Canonical

The tier-based model routing (docs/shared/agent-tiers.md) and explicit Task(subagent_type=..., model="haiku/sonnet/opus") dispatch is unique to this fork — the canonical project does not prescribe specific model tiers in this way.

09

Uniqueness

oh-my-codex (sigridjineth) — Uniqueness

Differs From Seeds

Derivative of oh-my-codex-yeachan; closest seeds are superpowers (skills-based) and spec-kit (multi-tier execution with explicit validation). The unique contribution is the most explicit multi-model tier routing in this batch: docs/shared/agent-tiers.md maps LOW/MEDIUM/HIGH complexity directly to Haiku/Sonnet/Opus, and Task(subagent_type=..., model="haiku") dispatch makes model selection deterministic rather than heuristic. Differs from taskmaster-ai (the other multi-model-aware seed) in routing by task complexity rather than by role (planner/executor). The LSP+AST tooling integration is not present in any other oh-my-codex variant or in the 11 seeds.

Positioning

  • Ambassador-fork with real value-add (not just a copy)
  • Most sophisticated model routing in the oh-my-codex family
  • Unique in adding Discord/Telegram notifications and LSP/AST tooling
  • Fewer stars (14) but more active than scalarian; maintained through March 2026
  • Listed as contributor and ambassador in the canonical project

Observable Failure Modes

  • No independent omx binary — relies on Codex plugin infrastructure
  • No quantitative ambiguity scoring (same gap as scalarian vs canonical)
  • docs/shared/agent-tiers.md file is a hard dependency for ultrawork routing; if absent, tier routing breaks
  • Codex plugin marketplace availability is required for preferred install path

Canonical Status

Not canonical; derivative but with attribution from the canonical project.

04

Workflow

oh-my-codex (sigridjineth) — Workflow

Canonical Entry Point

/oh-my-codex:omc-setup  → setup CODEX.md
/oh-my-codex:plan       → plan + architecture synthesis
/oh-my-codex:ultrawork  → parallel execution with tier routing
/oh-my-codex:review     → code review
/oh-my-codex:security-review  → security review
/oh-my-codex:code-review      → code review

Ultrawork Routing

$ultrawork
  ├── classify tasks by independence
  ├── route to tier:
  │   ├── LOW tier (Haiku) → simple lookups, definitions
  │   ├── MEDIUM tier (Sonnet) → standard implementation
  │   └── HIGH tier (Opus) → complex analysis, refactoring
  ├── fire independent tasks simultaneously
  ├── run dependent tasks sequentially
  └── background long operations (run_in_background: true)

Phases and Artifacts

Phase Skill Artifact Gate
Setup omc-setup CODEX.md none
Planning plan/ralplan Plan document User approval
Execution ultrawork/executor Code changes tier-based
Review code-review Review findings none
Verification (inline) Build/test output required

Tier Routing

Tier Model Use When
LOW Haiku Simple lookups, definitions, quick checks
MEDIUM Sonnet Standard implementation, normal changes
HIGH Opus Complex analysis, refactoring, architecture

Task dispatch:

  • Task(subagent_type="oh-my-codex:executor-low", model="haiku", ...)
  • Task(subagent_type="oh-my-codex:executor", model="sonnet", ...)
  • Task(subagent_type="oh-my-codex:executor-high", model="opus", ...)
06

Memory Context

oh-my-codex (sigridjineth) — Memory & Context

Session Memory

  • project-memory-session.mjs runs on every SessionStart — loads project memory into context
  • $writer-memory skill — session-scoped memory for writing workflows
  • $project-session-manager / $psm — drift detection and resumable sessions

State Storage

  • ~/.codex/ — global OMC install (hooks, agents, CODEX.md)
  • Per-session state: unspecified; presumed in ~/.codex/ or local .codex/

Hook-Based Context Priming

Every SessionStart (on startup/clear/compact) fires:

  1. session-start.mjs — initialize session state
  2. project-memory-session.mjs — load project memory

This ensures context is primed after compaction events.

Rate-Limit Recovery

Explicit tmux-based rate-limit wait/resume:

  • Agent detects API rate limit
  • Waits in tmux pane, resumes automatically when rate limit clears

Cross-Session Handoff

$project-session-manager handles drift detection and session resumability. Exact artifact format unknown — not documented in inspected files.

07

Orchestration

oh-my-codex (sigridjineth) — Orchestration

Multi-Agent: Yes

  • $ultrawork dispatches subagents via Task(subagent_type="oh-my-codex:executor-low/executor/executor-high")
  • $team enables tmux-based parallel workers
  • $swarm — N-agent coordinated execution with claim/checkpoint logic

Orchestration Pattern

parallel-fan-out (ultrawork fires independent tasks simultaneously) + hierarchical (team mode)

Multi-Model: Yes — most explicit in batch

Model tier routing:

Role Model Tier
Simple tasks claude-haiku (or Codex equivalent) LOW
Standard tasks claude-sonnet (or Codex equivalent) MEDIUM
Complex tasks claude-opus (or Codex equivalent) HIGH

This is the most explicit model-role mapping in the oh-my-codex family. The docs/shared/agent-tiers.md reference file defines the tier criteria.

Isolation Mechanism

process (each worker is a separate CLI subprocess)

Execution Mode

interactive-loop + one-shot (command invocation)

Consensus Mechanism

None

Prompt Chaining

Yes: keyword-detector → skill-injector → skill execution is a hook-driven chain

Context Compaction

Handled via project-memory-session.mjs on every SessionStart/compact trigger.

08

Ui Cli Surface

oh-my-codex (sigridjineth) — UI & CLI Surface

CLI

Minimal — node dist/cli/index.js install and doctor conflicts; no omx binary published independently.

Command surface within Codex (via plugin):

  • /oh-my-codex:omc-setup
  • /oh-my-codex:plan
  • /oh-my-codex:review
  • /oh-my-codex:ultrawork
  • /oh-my-codex:security-review
  • /oh-my-codex:code-review

HUD

HUD statusline support mentioned in README. Exact format: unknown.

Local Web Dashboard

None detected.

IDE Integration

None.

Notifications

  • $configure-discord — Discord webhook notifications
  • $configure-telegram — Telegram bot notifications

Unique in the oh-my-codex family — the canonical project uses a different notification system.

Developer Tooling

  • LSP tools: goto-definition, find-references, workspace symbols, diagnostics, prepare-rename, rename
  • AST tools: structural search and replace
  • These are skills/commands that invoke LSP/AST tooling, not MCP servers

Rate-Limit Recovery

tmux-based wait/resume for Codex API rate limits — built into the execution runtime.

Related frameworks

same archetype · same primary tool · same memory type

Claude-Flow / Ruflo ★ 55k

Eliminates single-agent context limits and sequential bottlenecks by orchestrating fault-tolerant swarms of specialized AI agents…

Hermes Agent (NousResearch) ★ 168k

Self-improving personal AI agent with closed learning loop, 7 terminal backends, and messaging gateway — not tied to any AI…

OpenCode ★ 165k

Terminal-first AI coding agent with multi-model routing, native desktop app, and a typed .opencode/ configuration system for…

OpenHands ★ 75k

Open-source AI software development platform (open-source Devin alternative) with Docker sandbox isolation, 77.6% SWE-bench…

DeerFlow ★ 70k

Long-horizon superagent that researches, codes, and creates by orchestrating parallel sub-agents with isolated contexts in Docker…

oh-my-openagent (omo) ★ 60k

Multi-provider AI agent orchestration for OpenCode: escape vendor lock-in by routing Sisyphus (Claude/Kimi/GLM) and Hephaestus…