yeachan-heo/oh-my-claudecode
40 skills19.4k installs2M starsGitHub
Install
npx skills add https://github.com/yeachan-heo/oh-my-claudecodeSkills in this repo
1Ai Slop Cleanerai-slop-cleaner is a level-3 skill from oh-my-claudecode for cleaning AI-generated code slop without changing intended behavior or drifting scope. It follows a bounded, deletion-first workflow with optional reviewer-only mode, targeting duplicate logic, dead code, wrapper layers, and weak tests left by prior agent passes. Triggers include explicit commands like deslop and anti-slop, plus requests to refactor noisy or over-abstracted implementations. Developers reach for ai-slop-cleaner when generated code passes tests but feels unmaintainable and needs surgical simplification before review or ship.901installs2RalplanRalplan is a shorthand alias for `/oh-my-claudecode:plan --consensus` in the oh-my-claudecode skill suite. It runs iterative planning where Planner, Architect, and Critic agents debate until consensus, using RALPLAN-DR structured deliberation—short mode by default and deliberate mode for high-risk work. Flags include `--interactive` for user prompts at decision points and `--architect codex` or `--critic codex` to route roles to specific models. Developers reach for ralplan when autopilot, ralph, or team-mode requests are too vague to execute safely and need a gated plan with explicit tradeoffs before any repository changes begin.850installs3UltraqaUltraQA is an oh-my-claudecode level-3 skill that activates autonomous QA cycling: qa-tester runs checks, an architect verifies results, fixes land, and the loop repeats until the goal is met. Flags include --tests, --build, --lint, --typecheck, --custom pattern, and --interactive. Conflict handling uses deterministic policies refuse, adopt_existing, and artifact_only instead of warnings. Developers reach for UltraQA after feature work when CI gates keep failing and they want the agent to iterate tests and fixes without manual babysitting.821installs4Hudhud is an oh-my-claudecode skill that configures the OMC Heads-Up Display on Claude Code's statusline. The skill supports quick commands such as /oh-my-claudecode:hud, setup, and presets including minimal, focused, and full layouts, writing configuration under ~/.claude while respecting CLAUDE_CONFIG_DIR overrides. Developers reach for hud when agent sessions feel noisy or when they want tighter real-time visibility into model status, tool activity, or session state. Setup can auto-repair missing HUD configuration and switch between display densities without editing JSON by hand. The skill is scoped as a config-writer at level 2 for ~/.claude paths, making it a focused ergonomics upgrade for daily Claude Code workflows.778installs5Deepinitdeepinit (level 4) performs deep codebase initialization by writing hierarchical AGENTS.md documentation across directories. Each file documents what a folder contains, component relationships, special instructions, and dependencies, with child files tagged <!-- Parent: ../AGENTS.md --> for navigable trees rooted at /AGENTS.md. Developers reach for deepinit when onboarding Claude Code or other agents to large monorepos, ensuring agents understand directory boundaries without re-scanning the full tree each session. The skill treats AGENTS.md as AI-readable documentation that links parent and child contexts for consistent agent behavior.672installs6Autopilotautopilot is a level-4 Claude Code skill from yeachan-heo/oh-my-claudecode that takes a 2-3 line product idea and autonomously handles the full lifecycle: requirements analysis, technical design, planning, parallel implementation, QA cycling, and multi-perspective validation. Developers trigger autopilot with phrases like autopilot, build me, create me, full auto, or handle it all when they want hands-off execution across planning, coding, testing, and validation phases. The skill explicitly avoids brainstorming-only sessions, single-file edits, or tasks where the developer wants to explore options interactively. autopilot produces working, verified code from minimal input by running multiple build subphases in sequence without manual handoffs between planning and implementation.660installs7Ultraworkultrawork is a level-4 parallel execution engine from oh-my-claudecode that routes independent work across multiple agents for faster throughput. The protocol emphasizes intent grounding, parallel context gathering, dependency-aware task graphs for non-trivial jobs, and concise evidence-backed execution summaries. ultrawork is a component, not a standalone persistence mode: it supplies parallelism and routing guidance without long-lived state, verification loops, or persistence management. Developers invoke ultrawork when several independent tasks can run simultaneously, often triggered by shortcuts like ulw or ultrawork in the prompt. Reach for it when wall-clock time dominates and tasks have clear independence boundaries.646installs8Ralphralph is a level-4 oh-my-claudecode skill that implements a self-referential persistence loop until every user story in prd.json has passes: true and reviewer verification. It wraps ultrawork's parallel execution with session persistence, automatic retry on failure, structured story tracking, and mandatory verification before a completion promise is accepted. Developers invoke ralph with optional flags --no-deslop, --critic=architect|critic|codex, plus a task description when complex work must be guaranteed complete. Arguments include configurable verification reviewers, and the loop continues across iterations until all PRD stories are verified, making it suited for multi-story agent tasks that one-shot prompts routinely leave incomplete.643installs9Omc Setupomc-setup is the canonical install and repair entry point for oh-my-claudecode (OMC) at skill level 2. Developers invoke it to install, refresh, or fix OMC after marketplace plugin installs, npm-based setups, or local-development configurations. The workflow executes immediately—agents run the documented steps rather than summarizing instructions—and respects the CLAUDE_CONFIG_DIR environment variable for ~/.claude paths. Choose omc-setup when OMC itself is missing, outdated, or broken rather than when debugging application code. After a successful run, downstream OMC automation is expected to operate without manual follow-up configuration.637installs10Deep Interviewdeep-interview is an oh-my-claudecode level-3 skill implementing Ouroboros-inspired Socratic questioning with mathematical ambiguity scoring. It exposes hidden assumptions through targeted questions, measures clarity across weighted dimensions, and blocks execution until ambiguity drops below a resolved threshold. Modes include --quick, --standard, and --deep, with optional --autoresearch, and output lands in .omc/specs/deep-interview-{slug}.md with approval-required handoff to planning. Developers reach for deep-interview when a feature request is too fuzzy for safe implementation and needs a gated spec before agents write code.622installs11Mcp Setupmcp-setup is a yeachan-heo/oh-my-claudecode level-2 skill that configures Model Context Protocol servers to extend Claude Code with external tools such as web search, filesystem access, and GitHub integration. It uses the `claude mcp add` CLI and guides setup through constrained AskUserQuestion flows—one question at a time with no more than three options per question, matching recent Claude Code parameter limits. Developers reach for mcp-setup when onboarding Claude Code agents and needing reliable MCP wiring instead of hand-editing config files. The skill focuses on popular, production-useful server paths rather than experimental MCP plugins.555installs12Visual VerdictVisual Verdict is a level-2 skill in yeachan-heo/oh-my-claudecode for structured visual QA. It accepts reference_images[], a generated_screenshot, and optional category_hint, then compares layout, spacing, typography, and component styling against intent. The skill returns deterministic pass/fail JSON guidance that can drive the next edit iteration before merge or release. Use it when visual fidelity requirements exist and you need automated judgment on polish and regressions instead of subjective chat feedback alone.551installs13Project Session ManagerProject Session Manager (PSM) is a worktree-first dev environment skill from oh-my-claudecode that exposes the `omc teleport` command for spinning up isolated git worktrees tied to issues, PRs, or feature names. Developers list active worktrees with `omc teleport list` and can layer tmux session orchestration on top when they want terminal multiplexing alongside branch isolation. The skill aliases to `psm` and targets Claude Code users who routinely juggle multiple branches, code reviews, and agent sessions in one repository. Reach for Project Session Manager when you need a repeatable, agent-driven workflow for parallel work instead of hand-rolling worktree and tmux scripts each time.545installs14Writer Memorywriter-memory is a yeachan-heo/oh-my-claudecode agent skill (level 7) that provides an agentic memory system for creative writers with first-class Korean storytelling workflow support. The skill tracks characters with emotional arcs, attitudes, dialogue tone, and speech levels, plus world settings, relationship dynamics, scenes, and themes across Claude sessions. Subcommands include init, char, rel, scene, query, validate, synopsis, status, and export for structured memory operations. Developers and technical authors reach for writer-memory when building agent-assisted fiction pipelines, maintaining series bible consistency, or exporting structured story context for collaborative writing tools.540installs15Teamteam is a skill from yeachan-heo/oh-my-claudecode that launches N coordinated Claude Code agents against one shared task list. Claude Code 2.1.178+ removed native TeamCreate and TeamDelete; with `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1`, each session exposes one implicit team and teammates spawn through the Agent tool with distinct `name` values. The skill preserves OMC legacy tmux and CLI worker orchestration documented via `omc team` and `/omc-teams`. Developers reach for it when a large task needs parallel agents that communicate and divide work. Argument hint format is `[N:agent-type] [ralph] <task description>`. Use it for complex multi-step coding efforts.536installs16Cancelcancel is the standard Oh My ClaudeCode (OMC) skill for intelligently detecting and exiting any active extended agent mode. It handles eight modes—autopilot, ralph, ultrawork, ultraqa, swarm, ultrapilot, pipeline, and team—and performs proper state cleanup when the stop hook signals work is complete. Developers invoke cancel to end long-running OMC sessions cleanly; if cancellation fails, --force forces teardown, and a 2-hour staleness timeout serves as last resort. The skill is level 2 in the OMC hierarchy and aliases cancel-ralph for ralph-specific exits.524installs17Learnerlearner is a yeachan-heo/oh-my-claudecode Level 7 self-improving skill that extracts learned expertise from the active conversation into reusable SKILL.md files. It maintains separate Expertise and Workflow sections, updating only Expertise during improvement cycles while keeping the extraction procedure stable. The skill is a deprecated compatibility alias—new workflows should use /oh-my-claudecode:skillify—but remains for internal implementation and history. Developers invoke learner when a debugging or implementation session surfaced durable principles worth packaging for future agent runs. Output encodes domain knowledge and stable workflows so agents repeat proven patterns instead of rediscovering fixes each session.524installs18Releaserelease is a yeachan-heo/oh-my-claudecode Level 3 generic release assistant invoked via /oh-my-claudecode:release with an optional version argument. On first run it inspects the project and CI to derive release rules, stores them in .omc/RELEASE_RULE.md, then walks through patch, minor, major, or explicit semver like 2.4.0 releases using those cached rules. Add --refresh to force re-analysis when pipelines change. Developers reach for release when a repo has non-obvious release conventions spanning changelogs, tags, and CI gates and they want a guided step-by-step ship flow instead of guessing commands each time. Subsequent runs reuse cached rules for faster, consistent version bumps.521installs19Skillskill is the oh-my-claudecode meta-skill—a CLI-like skill manager for local Claude agent skills. It scans three scopes: bundled built-in skills in the plugin `skills/` directory, user skills at `${CLAUDE_CONFIG_DIR:-~/.claude}/skills/omc-learned/`, and project skills at `.omc/skills/`, then parses YAML frontmatter and displays organized tables. Subcommands include list, add, remove, search, edit, and a setup wizard via `/skill <command> [args]`. Developers reach for this skill when their custom skill library grows and they need discovery, installation, and maintenance without manually editing directories. It is level 2 in the oh-my-claudecode hierarchy and targets Claude Code plugin users.519installs20Omc Doctoromc-doctor is the Oh My ClaudeCode Doctor skill (level 3) in the oh-my-claudecode repo. It inspects plugin cache directories under ~/.claude/plugins/cache/omc/oh-my-claudecode (honoring CLAUDE_CONFIG_DIR overrides), compares installed semver folders against latest releases via Node one-liners, and walks through fixes for common installation breakages. All path references respect CLAUDE_CONFIG_DIR when set, keeping diagnostics portable across custom config locations. Developers invoke omc-doctor when Claude Code fails to load OMC commands, plugin versions look stale, or post-upgrade hooks leave a partial install. The skill emphasizes repair steps over feature authoring.513installs21External Contextexternal-context is a yeachan-heo oh-my-claudecode level-4 skill that fetches external documentation, references, and technical context for a query by decomposing the topic into 2-5 facets and spawning parallel document-specialist Claude agents. Developers invoke it with /oh-my-claudecode:external-context followed by a search query or topic, such as JWT token rotation in Node.js, Prisma versus Drizzle for PostgreSQL, or latest React Server Components patterns. The skill fits advanced research moments when a single web search or manual doc reading is too narrow and multifaceted comparisons need concurrent specialist gathering. external-context returns consolidated reference material from outside the repository rather than editing local files. It complements codebase-aware skills by widening authoritative context before design or implementation choices. The skill assumes oh-my-claudecode orchestration and parallel agent spawning; it is not a lightweight inline snippet lookup.479installs22Tracetrace is an oh-my-claudecode level-2 skill that orchestrates the built-in tracer agent for ambiguous causal questions where the goal is explaining why an observed result happened. The workflow restates the observation, spawns 3 default tracer lanes in Claude team mode covering code-path, config or orchestration, and measurement or assumption mismatch causes, gathers evidence for and against each hypothesis in parallel, runs a rebuttal round, and ranks explanations on a 6-tier evidence strength hierarchy. Developers reach for trace on runtime bugs, latency regressions, architecture postmortems, and config routing mysteries where premature certainty would waste effort. Final output follows a 10-section synthesis contract including ranked hypotheses, critical unknown, and a single discriminating probe.471installs23Sciomcsciomc is a level 4 research skill in yeachan-heo/oh-my-claudecode that orchestrates parallel scientist agents for comprehensive analysis with optional AUTO mode for fully autonomous execution. The workflow runs four stages: decomposition of the research goal into independent stages or hypotheses, parallel agent execution on each stage, verification through cross-validation and consistency checks, and synthesis into a comprehensive report. Developers reach for sciomc when a question spans multiple domains, requires competing hypotheses tested in parallel, or when a single agent pass would miss contradictions in source material. AUTO mode enables hands-off runs after the research goal is supplied via the argument-hint research goal parameter. Catalog metadata records 402 installs. Use sciomc for literature-style technical investigations, comparative architecture studies, and evidence-heavy decision memos before committing to build plans. The level 4 skill fits comparative technology evaluations, market scans, and architecture trade-off studies where verified citations matter more than a single-pass summary from one agent.467installs24Ccgccg is a Claude-Codex-Gemini tri-model orchestration skill from yeachan-heo/oh-my-claudecode that routes through the canonical /ask skill with /ask codex and /ask antigravity (or Gemini), then has Claude synthesize both outputs into one answer. Developers invoke it for parallel external perspectives on code review, architecture validation, and cross-domain requests combining backend analysis with frontend or UX feedback without launching tmux team workers. CCG is rated level 5 and suits fast advisor-style parallel input when Codex and Antigravity may disagree. Reach for ccg when multi-model cross-validation is needed in a single request.461installs25Configure Notificationsconfigure-notifications is a level-2 skill from yeachan-heo/oh-my-claudecode that configures external notification integrations for oh-my-claudecode through natural-language setup flows. It supports three providers: Telegram via bot token and chat ID, Discord via webhook or bot token plus channel ID, and Slack via incoming webhook URL. The skill detects existing settings in .omc-config.json under the Claude config directory, routes to the correct provider based on trigger phrases like configure telegram or slack webhook, and walks users interactively through BotFather bot creation, chat ID discovery, and webhook validation. Developers reach for configure-notifications when they want mobile or desktop alerts for Claude Code session completion, input requests, or background task results without polling the terminal. Output is a validated .omc-config.json notifications block with provider credentials and parse mode settings. The skill does not build custom notification servers or general application push notification systems outside the OMC Claude Code integration.452installs26Omc Teamsomc-teams is a yeachan-heo/oh-my-claudecode level-4 skill for CLI-first parallel agent execution. It spawns N worker processes in tmux panes via `/oh-my-claudecode:omc-teams N:claude "task"` syntax, supporting agent types claude, codex, gemini, antigravity, grok, and cursor (cursor workers are executor-style only). The skill is legacy compatibility for the CLI runtime—use `omc team ...` commands rather than deprecated MCP runtime tools. Developers reach for omc-teams when independent tasks can run simultaneously across heterogeneous CLI agents instead of serial single-process execution.451installs27Askask is a skill in yeachan-heo/oh-my-claudecode that implements process-first advisor routing via the `omc ask` command. Developers specify an advisor—claude, codex, gemini, antigravity, grok, or cursor—and a question; OMC handles CLI invocation and captures the reply as a durable ask artifact instead of raw terminal output. Typical prompts include security patch review, UX flow suggestions, and cross-model planning checks. The skill eliminates hand-assembled CLI flags and standardizes artifact storage inside the OMC framework. Use ask when you want a Codex security pass, a Gemini UX critique, or a Cursor-specific review without leaving your Claude Code session or losing the response.400installs28Setupsetup is a level-2 oh-my-claudecode skill that exposes `/oh-my-claudecode:setup` as the single configuration entrypoint for Claude Code sessions. The skill inspects only the first CLI argument and delegates to one of three flows: a full setup wizard (`omc-setup`), installation diagnostics (`omc-doctor`), or MCP server configuration (`mcp-setup`). Supported invocations include `/oh-my-claudecode:setup`, `/oh-my-claudecode:setup doctor`, `/oh-my-claudecode:setup mcp`, and `/oh-my-claudecode:setup wizard --local`, while legacy slash commands remain for compatibility. Developers reach for setup when onboarding a new machine, repairing a broken OMC install, or wiring MCP servers before running multi-agent workflows. The routing design prevents misrouted setup questions and keeps documentation aligned on one canonical command instead of three separate entrypoints. Terminal users can also run `omc setup` or `omc doctor` from the npm CLI path after installing oh-my-claudecode.388installs29Deep Divedeep-dive is an oh-my-claudecode orchestration skill with a 2-stage pipeline: trace investigates why something happened, then deep-interview crystallizes what to do about it. The trace stage launches 3 parallel causal investigation lanes whose findings feed a 3-point injection into the interview stage, producing a handoff spec at .omc/specs/deep-dive-{slug}.md. Developers reach for deep-dive when problems are ambiguous, regressions lack obvious owners, or requirements need evidence-backed definition before the plan and autopilot skills run. Triggers include phrases like deep dive, trace and interview, and investigate deeply, with next-skill set to plan using --consensus --direct.387installs30Self Improveself-improve is a fully autonomous code improvement loop that runs setup, research, planning, execution, tournament selection, history tracking, visualization, and stop-condition checks. self-improve is designed for developers who want an agent to repeatedly propose and apply improvements, compare candidate changes, and preserve a history of iterations rather than doing one-off refactors. self-improve is most useful when a codebase needs ongoing quality and maintainability improvements, especially when improvements must be selected among alternatives instead of applied blindly. Developers reach for self-improve when they want a structured, repeatable loop that can plan, execute, evaluate outcomes, and stop based on explicit criteria.382installs31Omc Referenceomc-reference is a built-in reference skill from yeachan-heo/oh-my-claudecode for the OMC agent framework. The skill auto-loads when delegating to agents, using OMC tools, orchestrating teams, making commits, or invoking skills, and exposes the catalog under the oh-my-claudecode: prefix. Documented agents include explore on haiku for fast codebase search, analyst on opus for requirements clarity, planner on opus for execution sequencing, and architect on opus for system design, with full prompts stored in agents/*.md files. Developers reach for omc-reference when they need detailed routing, commit protocol, or skills registry information without pasting the entire OMC documentation into CLAUDE.md.354installs32Verifyverify is a completion-gate skill from yeachan-heo/oh-my-claudecode that replaces vague it-should-work claims with concrete evidence. The SKILL.md defines a five-step workflow: identify the exact behavior to prove, prefer existing tests first, run the narrowest direct verification command if coverage is missing, gather manual validation steps when automation is insufficient, and report only what was actually verified. The verification order runs existing tests, then typecheck or build, then narrow direct commands, then manual checks. Developers and agents invoke verify when a user wants confidence that a feature, fix, or refactor truly works before marking a task done or opening a pull request.311installs33Skillifyskillify is an Oh My Claudecode (OMC) meta-skill—also invokable as the deprecated alias `/oh-my-claudecode:learner`—that turns a repeatable agent session into a structured skill draft with YAML frontmatter, triggers, ordered steps, success criteria, and pitfalls. Before extraction, a three-part quality gate asks whether the workflow is Googleable in five minutes, specific to the codebase, and costly to rediscover through debugging or operations. skillify outputs complete SKILL.md files to `${CLAUDE_CONFIG_DIR}/skills/omc-learned/` or `.omc/skills/`, deciding between repo built-in skills, user learned skills, or documentation-only. Use skillify after a session solves a non-obvious, project-specific problem you expect to repeat, not for generic library tutorials or boilerplate snippets.310installs34Rememberremember is an oh-my-claudecode skill that reviews reusable knowledge from a coding session and routes each item to the correct memory surface. It distinguishes four destinations: project memory for durable team facts, notepad priority for high-signal next-turn context, notepad working for temporary session notes, and Docs or AGENTS or CLAUDE files for lasting conventions. The workflow gathers findings, classifies them as durable facts, working notes, operator preferences, or stale duplicates, proposes destinations, writes only appropriate stores, and flags conflicts. Rules forbid dumping everything into one store and require concise, actionable entries marked uncertain when unverified. Developers reach for remember when a session uncovered architecture decisions, conventions, or preferences that should survive the next conversation instead of being re-discovered.303installs35Debugdebug from oh-my-claudecode helps developers diagnose current OMC or Claude Code session problems, workflow breakage, and confusing runtime behavior. The skill follows five steps: read the issue, inspect local evidence from trace tools, state tools, notepad or project memory, and failing tests or commands, reproduce narrowly, separate symptoms from root cause, and recommend the smallest corrective action. Use debug when agents loop, hooks misfire, traces show silent failures, or project memory disagrees with runtime state. It prioritizes local artifacts—logs, traces, state dumps—over speculative refactors so engineers get a concrete next fix fast.301installs36WikiWiki is a persistent, self-maintained markdown knowledge base skill for Claude Code that lets agents ingest and query project knowledge across sessions. Builders use it to capture decisions, patterns, and context so they are not rediscovered from scratch on the next session. It supports multi-page ingestion and tag-based search with cited snippet responses.295installs37AutoresearchAutoresearch is a Claude Code skill for persistent, mission-focused automated experimentation. It maintains durable logs of research iterations and runs scheduled evaluation cycles via Claude Code cron, requiring a mission and evaluator obtained from the deep-interview workflow. Solo builders use it when they want to continuously improve a specific output or behavior without manually rerunning experiments each time.257installs38Ultragoalultragoal is a persistent project-tracking skill for the oh-my-claudecode framework that stores a durable goal ledger directly in the repository. It breaks large work into ordered stories with attempt tracking and evidence logs, ensuring nothing is lost when Claude sessions restart. Solo builders reach for it when tackling projects large enough to span multiple working sessions and needing reliable continuity with a structured verification workflow before marking work complete.183installs39Local Build Reminderlocal-build-reminder is a Claude Code skill for ai & agent building. It helps solo builders move faster with AI-assisted coding.100installs40Planplan is a Claude Code skill for ai & agent building. It helps solo builders move faster with AI-assisted coding.70installs