oh-my-claudecode | Ry Walker Research

Key takeaways

36.2K stars and 3.3K forks in roughly five months — the repo was created January 9, 2026 and shipped 232 releases by v4.14.6 (June 9, 2026), one of the fastest star trajectories in the Claude Code ecosystem
"Team" is the canonical surface: a staged team-plan → team-prd → team-exec → team-verify → team-fix loop inside Claude Code, plus an `omc team` CLI that spawns real Claude/Codex/Gemini/Grok processes in tmux panes
Pitch is "zero learning curve" — natural-language intake (/autopilot, /deep-interview), 32 specialized agents, 40+ skills, and model routing the project claims saves 30-50% on tokens
Single-maintainer concentration and extreme release velocity cut both ways: breaking changes land fast (swarm removed in v4.1.7, Codex/Gemini MCP servers removed in v4.4.0), and English-language community discussion is thin relative to the star count

FAQ

What is oh-my-claudecode?

oh-my-claudecode (OMC) is an open-source MIT-licensed orchestration framework that layers multi-agent teams, execution modes, and specialized agent roles on top of Claude Code, installable as a Claude Code plugin or an npm CLI.

How much does oh-my-claudecode cost?

The software is free under MIT; you pay for the underlying model usage — a Claude subscription or API key, plus optional Codex, Gemini, and Grok CLI access for multi-model workers.

How does oh-my-claudecode run multiple agents in parallel?

Two runtimes: an in-session native-team workflow (/team) that runs a staged plan → PRD → execute → verify → fix pipeline inside Claude Code, and an `omc team` CLI that spawns on-demand Claude, Codex, Gemini, or Grok CLI workers in tmux split-panes that die when their task completes.

How is oh-my-claudecode different from Gastown?

Gastown is a Go-based town-of-agents orchestrator with its own work-tracking substrate (Beads); OMC stays inside Claude Code itself as a plugin, betting on pre-defined agents, skills, and natural-language modes rather than an external control plane.

Executive Summary

oh-my-claudecode (OMC) bills itself as "multi-agent orchestration for Claude Code. Zero learning curve" — and, on its own site, as "a weapon, not a tool."^[1]^[2] It installs as a Claude Code plugin (or an npm CLI) and replaces the single chat loop with a teams-first workflow: describe the task in natural language, and OMC runs a staged pipeline — team-plan → team-prd → team-exec → team-verify → team-fix — looping the fix stage until verification passes.^[1] A separate omc team CLI spawns real Claude, Codex, Gemini, or Grok CLI processes as tmux split-pane workers that spin up on demand and die when their task completes.^[1]

The trajectory is the headline. The repo was created January 9, 2026 and reached 36.2K stars and 3.3K forks by June 11, 2026 — five months — with 232 releases shipped along the way, landing at v4.14.6 on June 9.^[3]^[4] Creator and lead Yeachan Heo is a Seoul-based algorithmic trader who leads Quant.start(), which he describes as Korea's biggest quant-trading community; OMC's README ships in seven languages, and the project's center of gravity is its Discord and a largely Korean contributor base rather than Hacker News or Reddit.^[5]^[1]^[6]

Attribute	Value
Creator	Yeachan Heo ("Bellman"), Seoul; leader of Quant.start() quant-trading community^[5]
First commit	January 9, 2026^[3]
GitHub Stars	36.2K (3.3K forks) as of June 2026^[3]
Releases	232; latest v4.14.6 (June 9, 2026)^[4]
License	MIT^[3]
Language	TypeScript^[3]
Funding	None disclosed; GitHub Sponsors solicited in README^[1]

Product Overview

The pitch is anti-configuration: "Don't learn Claude Code. Just use OMC."^[1] After install, /autopilot "build a REST API for managing tasks" is the whole interface — OMC decides decomposition, agent assignment, and model routing. For vague requirements, /deep-interview runs Socratic questioning to expose hidden assumptions before any code is written.^[1] Since v4.1.7, Team is the canonical orchestration surface (the legacy swarm keyword was removed), and native teams require Claude Code's experimental CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS flag — OMC falls back to non-team execution when it is off.^[1]

Under the hood it claims 32 specialized agents (architect, executor, designer, QA, security auditor, data scientist, and so on) and 40+ skills, each role matched to a model tier, with smart routing the project says saves 30-50% on tokens; an HUD statusline shows what the orchestrator is doing in real time, and a learning layer extracts problem-solving patterns from sessions for reuse.^[7]^[1]

Key Capabilities

Capability	Description
Team pipeline	Staged `team-plan → team-prd → team-exec → team-verify → team-fix` loop; persists until verified complete^[1]
tmux CLI workers	`omc team N:codex\|gemini\|grok\|claude "..."` spawns N real CLI processes in tmux panes; on-demand, self-terminating^[1]
Execution modes	Autopilot, Ultrawork, Ralph, Ultrapilot (3-5 parallel agents), Ecomode, Pipeline^[7]^[1]
Specialized agents	32 agents and 40+ skills, model-tiered per task complexity^[7]
Deep Interview	Socratic requirement clarification before execution; absorbed a community convergence-gate pattern in v4.6.0^[1]^[6]
Multi-model advisors	`/ask codex\|gemini\|grok` and `/ccg` tri-model synthesis (Codex + Gemini review, Claude synthesizes)^[1]
Cost routing	Vendor-claimed 30-50% token savings via model routing^[1]

Product Surfaces

Surface	Description	Availability
Claude Code plugin	`/plugin marketplace add` + `/plugin install oh-my-claudecode`; in-session skills (`/autopilot`, `/team`, …)	GA^[1]
npm CLI	`npm i -g oh-my-claude-sisyphus` — installs `omc` and `oh-my-claudecode` commands; tmux team runtime	GA^[8]^[1]
Docs site + Discord	Hosted docs, CLI reference, migration guide; community Discord	GA^[1]

Technical Architecture

OMC is a TypeScript layer over Claude Code's native extension points — per independent analysis it is "built on Claude Code's native Hooks (shell scripts) to avoid ToS violations," a deliberate contrast with harnesses that re-implement the client against Anthropic's subscription terms.^[7] The in-session runtime rides Claude Code's experimental native agent teams; the terminal runtime drives external CLIs — including OpenAI's Codex CLI, Gemini CLI, and Grok — as tmux panes, requiring those CLIs and their auth to be installed separately.^[1] v4.4.0 removed the earlier Codex/Gemini MCP-server providers in favor of this CLI-first model.^[1]

# Plugin path (recommended), one line at a time inside Claude Code
/plugin marketplace add https://github.com/Yeachan-Heo/oh-my-claudecode
/plugin install oh-my-claudecode

# npm path — note the package name differs from the brand
npm i -g oh-my-claude-sisyphus@latest
omc setup

Key Technical Details

Aspect	Detail
Deployment	Local/self-hosted plugin + CLI; no hosted service^[1]
Models	Claude (primary) plus Codex, Gemini, Grok CLI workers; model-tier routing per agent role^[1]
Dependencies	Claude Code; tmux + external CLIs for multi-model teams; `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1` for native teams^[1]
Open Source	MIT, single repo; sibling project oh-my-codex targets Codex CLI^[3]^[1]

Strengths

Category-defining adoption speed — 36.2K stars and 3.3K forks in five months from a January 2026 first commit, with an active Discord and READMEs in seven languages.^[3]^[1]
Two genuinely different runtimes — in-session native teams for coordinated Claude agents, and tmux CLI workers for real Codex/Gemini/Grok processes; most rivals offer only one of these.^[1]
Verification is built into the loop — the team pipeline ends in team-verify → team-fix cycles rather than declaring victory at first diff; the README's framing is "won't give up until the job is verified complete."^[1]
ToS-conscious architecture — building on Claude Code hooks rather than reverse-engineering the client avoids the subscription-blocking fate that hit oh-my-opencode-style harnesses.^[7]
Relentless iteration — 232 releases in five months, with community patterns (the Deep Interview convergence gates) absorbed into core within weeks.^[4]^[6]

Cautions

Effectively a single-maintainer project — Yeachan Heo is the sole listed core maintainer; the top external collaborator has 65 commits. The bus factor is one, and no foundation or company stands behind it.^[1]^[5]
Velocity means breaking churn — swarm removed in v4.1.7, Codex/Gemini MCP servers removed in v4.4.0, an autoresearch command hard-deprecated, and a migration guide maintained because upgrades regularly need one.^[1]
Naming and packaging confusion — the brand is oh-my-claudecode but the npm package is oh-my-claude-sisyphus, a documented gotcha the README itself has to explain twice.^[1]^[8]
Rides an experimental upstream flag — native team mode depends on CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS; if Anthropic changes or ships its own equivalent, OMC's canonical surface is exposed. Independent commentary flags exactly this: community tools often stall once official features catch up.^[1]^[7]
Token-cost claims are vendor-stated — the 30-50% savings figure is the project's own; parallel multi-agent execution can just as easily multiply spend, and multi-model teams require separate Codex/Gemini/Grok subscriptions.^[1]^[9]
English-language community signal is thin for the star count — no dedicated HN launch thread, near-zero Reddit footprint, and only 3 open issues against 36K stars as of June 2026; sentiment lives in Discord and the Korean dev community, which is harder to audit independently.^[6]^[3]

What Developers Say

Independent English-language discussion is sparse relative to 36.2K stars as of June 2026 — no HN front-page thread exists, and the substantive third-party commentary comes from individual blog reviews; that gap is itself a data point.^[6]

"oh-my-claudecode's pre-defined agents and execution modes are ahead in terms of convenience… the Ultrapilot mode in particular could run 3-5 agents in parallel, dramatically speeding up work." — Jeongil Jeong, developer blogger^[7]

"Community tools often stop being updated once official features catch up." — Jeongil Jeong, on the risk Anthropic's native Agent Teams poses to OMC^[7]

"The product surface is large and opinionated, so it can feel like a lot if you only wanted a tiny Claude Code tweak instead of a full workflow layer." — Vibe Coding Hub review^[9]

"The 'Deep Interview' pattern from this project was recently merged into the oh-my-claudecode (OMC) v4.6.0 release as an official skill." — q00 on Hacker News, a contributor whose convergence-gate pattern OMC absorbed^[6]

Pricing & Licensing

Tier	Price	Includes
Open source	Free	Full framework — plugin, CLI, all agents, modes, and skills^[3]
Sponsorship	Optional	GitHub Sponsors link in README; no paid tier or hosted offering^[1]

Licensing model: MIT, single repo under the maintainer's personal GitHub account.^[3]

Hidden costs: All model usage — a Claude subscription or API key is the floor, and multi-model teams add Codex, Gemini, and/or Grok CLI subscriptions on top; parallel workers multiply token burn, and tmux plus each provider's CLI must be installed and authenticated separately.^[1]^[9]

Competitive Positioning

Direct Competitors

Competitor	Differentiation
Gastown	Steve Yegge's Go orchestrator runs 20-30 mixed agents with its own Beads work-tracking substrate and a hosted option; OMC stays inside Claude Code as a plugin with no external control plane
Metaswarm	Dave Sifry's 18-agent, 9-phase pipeline leads with cross-model adversarial review; OMC offers similar multi-model checking (/ccg) but optimizes for zero-config natural-language intake
Symphony	OpenAI's issue-tracker-driven orchestrator (issues in, PRs out) that OpenAI won't maintain as a product; OMC is interactively driven and aggressively maintained
Claude Code native Agent Teams	Anthropic's first-party teams compose dynamically per task; OMC's bet is that pre-defined agents, modes, and verify-fix loops beat on-the-fly composition^[7]

When to Choose oh-my-claudecode Over Alternatives

Choose OMC when: you live in Claude Code, want orchestration with no harness to learn or host, and value pre-built agents, verify-fix persistence, and optional Codex/Gemini/Grok workers from one install.
Choose Gastown when: you want an external control plane with durable work tracking across many agents and repos, or a hosted option.
Choose Metaswarm when: enforced quality gates and cross-model adversarial review are the priority over convenience.
Choose Symphony when: you want issue-driven, fully autonomous flow and accept an unmaintained reference implementation.

Ideal Customer Profile

Best fit:

Individual developers and small teams already paying for Claude Code who want multi-agent parallelism without building or hosting a harness
Developers running multi-provider subscriptions (Claude + Codex/Gemini/Grok) who want one orchestration surface across all of them
Terminal-native users comfortable with tmux and fast-moving tooling

Poor fit:

Enterprises needing vendor accountability, SLAs, or a maintainer organization larger than one person
Teams that want issue-tracker-driven autonomy rather than interactive session orchestration
Anyone allergic to breaking changes — 232 releases in five months means the ground moves^[4]

Viability Assessment

Factor	Assessment
Financial Health	N/A — unfunded MIT project sustained by one maintainer plus GitHub Sponsors^[1]
Market Position	Strongest star count in the Claude Code orchestration niche — 36.2K vs. Gastown's 15.9K — five months after first commit^[3]
Innovation Pace	Extreme — 232 releases to v4.14.6 in five months; community patterns merged within weeks^[4]^[6]
Community/Ecosystem	Large but lopsided — 3.3K forks, active Discord, seven README languages, yet minimal HN/Reddit footprint and a heavily Korea-centered contributor base^[3]^[1]^[6]
Long-term Outlook	Hinges on two risks: the bus factor of one, and Anthropic's native Agent Teams subsuming the value proposition^[1]^[7]

The growth is real and the architecture is sensibly parasitic — by building on Claude Code's official hooks and experimental teams rather than against Anthropic's terms, OMC positions itself as an accelerant to the platform rather than a competitor to it.^[7] But that same positioning is the existential risk: everything OMC does is a candidate for first-party absorption, and the project's survival depends on one Seoul-based maintainer out-shipping Anthropic's roadmap indefinitely.^[5]^[1]

Bottom Line

oh-my-claudecode is the highest-adoption, fastest-moving way to turn a single Claude Code session into a verified multi-agent team today — free, MIT-licensed, ToS-conscious, and genuinely multi-model via tmux workers. The trade is structural: a bus factor of one, breaking changes as a way of life, vendor-stated cost claims, and a community whose sentiment is hard to audit from outside its Discord.

Recommended for: Claude Code power users and small teams who want zero-config multi-agent orchestration, verify-fix persistence, and optional Codex/Gemini/Grok workers without hosting anything.

Not recommended for: Enterprises needing maintainer accountability, teams wanting issue-driven autonomy over interactive sessions, or anyone who can't absorb frequent breaking changes.

Outlook: Watch whether Anthropic's native Agent Teams (which OMC's canonical mode already depends on, behind an experimental flag) absorbs the category; whether maintainership broadens beyond one person; and whether the star-to-discussion gap closes with substantive independent reviews.

Research by Ry Walker Research • methodology

Sources