oh-my-claude

7 Custom Agents

11 Built-in Skills

18 Lifecycle Hooks

1 Plugin Install

Three steps. Zero hand-holding.

You describe the work. Agents handle the rest. Your context window stays clean.

STEP 1

You ask

Describe what you need. A feature, a bug fix, a refactor — whatever. Type it like you're talking to a senior dev.

$ "Add auth to the API routes"

STEP 2

Agents swarm

Built-in Explore finds files. Librarian reads them. Plan mode designs. Tasks implement. Your context window never sees the raw dump.

$ Explore → librarian → Plan → Task

STEP 3

You get results

Clean output lands in your session. Code written, tests passing, files changed. You stayed in reasoning mode the whole time.

$ 4 files changed, tests passing

Plan mode, supercharged.

Enter plan mode and get a structured 7-step pipeline automatically. No magic words needed — just Shift+Tab.

Recon

Explore and librarian gather codebase context before any questions are asked.

Explore + librarian

Interview

Smart, recon-informed questions via AskUserQuestion. No guessing.

Research

Targeted deep research informed by your answers.

Explore + librarian

Gap Analysis

Advisor catches hidden assumptions, missing context, and scope risks.

advisor

Risk Pass

Risk-assessor evaluates cross-stack change risk before plan approval.

risk-assessor

Write Plan

Structured plan with file:line refs, decisions, and risks.

Critic Review

Critic must approve before plan exits. Loops until the plan is solid.

critic

Want more? Say the word.

Drop a keyword in your prompt to activate specialized execution modes.

ultrawork

ulw

Maximum execution intensity. Relentless task tracking and parallel delegation.

ultraresearch

ulr

Systematic research with parallel searches, cross-referencing, and citations.

ultradebug

uld

Forensic debugging protocol. Multiple hypotheses, 3-strike escalation.

Your context is not a dumpster

Four systems that keep your session focused on what matters: reasoning.

Your context is sacred

Raw file dumps destroy response quality. Subagents handle reading, searching, and grepping — your session only sees the conclusions.

Structured execution, not vibes

Automatic task creation, progress tracking, and parallel agent delegation. Every complex task gets a plan before anyone writes code.

Git workflows that don't suck

Conventional commits, draft PRs, worktree isolation. Built-in skills handle the ceremony so you focus on the code.

Hooks that enforce your standards

Python hooks inject context, validate actions, and catch mistakes automatically. Your patterns, enforced on every interaction.

Built for how you actually work

Real workflows, not marketing demos.

Exploring a New Codebase

Drop into an unfamiliar repo without flooding your context with every file.

Explore finds → 47 files matched

librarian reads → 3 relevant modules

Plan mode → dependency graph built

> "Auth lives in src/middleware/, uses JWT"

Implementing a Feature

Describe what you want. The agent swarm handles the rest while you review.

Plan mode → 5-step plan created

Task[1] → src/routes/auth.ts written

Task[2] → src/middleware/jwt.ts written

validator → all 12 tests passing

Debugging Production

When you're stuck after two attempts, use ultradebug skill for deep reasoning you shouldn't burn context on.

ultradebug → root cause: race condition

in useEffect cleanup on unmount

Task → fix applied to 2 files

validator → regression suite green

Specialized Agents

7 custom agents that extend Claude Code's built-in capabilities. Your context window stays clean while they do the heavy lifting.

Custom Agents

librarian

read

Smart file reading agent. Reads intelligently, summarizes large content, extracts relevant sections.

Use: Reading large files, Git analysis, summarizing code

advisor

plan

Pre-planning gap analysis. Reviews requirements for hidden assumptions, missing context, and scope risks BEFORE plan generation.

Use: Gap analysis before writing plans, catching scope risks

risk-assessor

assess

Proactive change-risk agent. Assesses functional, security, compatibility, performance, and operational risk for plans and diffs.

Use: Non-trivial planning risk pass, PR/diff risk triage

critic

review

Critical plan reviewer. Reviews plans to find flaws, edge cases, problems BEFORE execution.

Use: Reviewing implementation plans

validator

verify

Quality assurance agent. Runs tests, linters, type checks. Reports pass/fail with specific errors.

Use: Running test suites, CI validation

code-reviewer

review

Post-implementation review agent. Reviews code against requirements and coding standards before merge.

Use: Code review after implementation, quality checks before merge

security-auditor

audit

Security-focused code review. Finds OWASP vulnerabilities, hardcoded secrets, injection flaws, and analyzes dependency lock-files.

Use: Security audits, pre-merge security review, dependency CVE analysis

Built-in Integration

oh-my-claude leverages Claude Code's native agents for common tasks:

Explore

Claude Code built-in. Finds files, locates definitions, maps dependencies. Automatic codebase reconnaissance.

Use: File discovery, definition lookup, pattern matching

Plan

plan

Claude Code built-in. Structured planning mode with task decomposition and parallel execution strategies.

Use: Multi-step implementations, complex workflows

Task

implement

Claude Code built-in. General-purpose implementation agent. Executes specific tasks completely.

Use: Writing code, making edits, creating files

One lead. Many teammates.

Claude Code's agent teams let you spawn parallel sessions that coordinate through shared task lists and messaging. oh-my-claude makes every session smarter.

You describe

A multi-module feature, a cross-layer refactor, parallel research — any task that benefits from concurrent work.

Team spawns

Teammates launch as independent sessions, each with their own context window. Shared task list keeps everyone aligned.

Specialists assist

Both leads and teammates tap specialist agents — librarian reads, advisor and risk-assessor evaluate plans, critic reviews, validator runs checks.

Work converges

Tasks complete in parallel. Hooks enforce standards across every session. You review results, not process.

What the plugin adds

Native teams handle coordination. oh-my-claude handles quality.

Context protection everywhere

Every teammate gets the same context discipline. Hooks enforce delegation to specialists and warn on context pressure — automatically.

Specialist agents for every role

Teammates spawn librarian for file reading, advisor for gap analysis, risk-assessor for change risk, and critic for plan review. No teammate wastes context on raw file dumps.

Standards enforced at scale

Commit quality, TDD requirements, danger blocking, and safe permissions — all hooks fire in every session. Three teammates means three times the enforcement, not three times the risk.

Dual-mode intelligence

Hooks detect whether they are running in a lead, teammate, or solo session and adapt behavior accordingly. Team leads get delegation reminders. Teammates get implementation focus.

Three commands. Zero config.

Install the plugin. Start a new session. That's it.

$ curl -LsSf https://astral.sh/uv/install.sh | sh
$ claude plugin marketplace add techdufus/oh-my-claude
$ claude plugin install oh-my-claude@oh-my-claude
→ Plugin installed. Context protection active.