This week's Breakout riser in Claude Code skill searches is "oh my claude code". The query maps to an entire configuration framework: oh-my-claudecode, a 38,017-star repo whose pitch is multi-agent orchestration with zero learning curve. Its motto: "Don't learn Claude Code. Just use OMC." Its most-installed skill, at 845 installs, is a cleanup guardrail called ai-slop-cleaner; the headline autopilot ranks sixth.
Every successful tool eventually grows this layer. zsh existed for nearly two decades before oh-my-zsh turned its configuration surface into a one-line install. Vim got vim-plug; Neovim got whole distributions like LazyVim. The pattern repeats because the base tool's power outgrows the average user's willingness to configure it, and someone packages taste. Claude Code crossed that line in well under two years.
What OMC is
A plugin that installs a stack of orchestration skills on top of Claude Code: planning loops where multiple agents debate, QA cycles that rerun until checks pass, a persistence loop that keeps working until every story in a PRD passes, and an autopilot that claims to take a 2-3 line product idea to shipped code. The project's own website calls it "a weapon, not a tool."
What developers actually install from it
We track installs per skill across the catalog of Claude Code agent skills. The ranking inside OMC inverts its own marketing:
| Rank | Skill | Installs | What it does |
|---|---|---|---|
| 1 | ai-slop-cleaner (GitHub) | 845 | Deletion-first cleanup of AI-generated code, scoped to never change intended behavior |
| 2 | ralplan (GitHub) | 824 | Planning by consensus: Planner, Architect and Critic agents debate until they agree |
| 3 | ultraqa (GitHub) | 805 | Autonomous QA cycling: test, verify, fix, repeat until the goal passes |
| 4 | hud (GitHub) | 775 | A statusline heads-up display showing what the orchestrator is doing |
| 5 | deepinit (GitHub) | 664 | Writes hierarchical AGENTS.md documentation across your repo |
| 6 | autopilot (GitHub) | 653 | Full lifecycle from a short product idea, autonomously |
| 7 | ralph (GitHub) | 636 | Self-referential persistence loop until every PRD story passes verification |
The three most-installed skills are the bounded ones: clean up slop without changing behavior, argue a plan to consensus before touching code, loop QA against real checks. The two headline autonomy features, autopilot and ralph, sit at the bottom of the table.
The gap is narrow (845 vs 636, and these numbers move daily), so don't read it as a chasm. But the ordering agrees with what we have measured in recent weeks: the fastest-growing skill searches are restraint layers (token diets, taste constraints), and now, inside the most maximalist framework in the ecosystem, the guardrail skills out-install the weapon.
The framework layer is spreading to Codex
The same author ships oh-my-codex for OpenAI's Codex CLI, and "codex-plugin-cc" is another fast-rising search this week (+80%) as Codex users go looking for Claude Code's plugin mechanics on their side of the fence. Configuration frameworks are becoming portable across agents, the way oh-my-zsh themes outlived any single shell setup.
If you try it
The install data suggests a sane order: start with ai-slop-cleaner and ralplan (bounded, easy to evaluate on a single diff), keep ultraqa pointed at a real test suite, and treat autopilot and ralph as demos until you have seen what the bounded skills do to your codebase.
Data: live catalog, refreshed daily from skills.sh, GitHub and MCP registries, ranked by real installs.
Skillselion is a directory of Claude Code, Codex and Cursor extensions ranked by real installs. We are not affiliated with Anthropic, OpenAI or Cursor.
Top comments (0)