How Skill Composition Layering Works in Oh-My-ClaudeCode: Execution, Enhancement, and Guarantee

Oh-My-ClaudeCode implements skill composition layering as a three-tier wrapper system where execution skills handle core tasks, enhancement skills add capabilities like parallelism or git automation, and guarantee skills enforce verification criteria before completion.

Oh-My-ClaudeCode (OMC) eliminates agent switching by injecting skills that wrap the core execution flow through a deterministic layering architecture. This skill composition layering approach processes every task through three logical tiers—execution, enhancement, and guarantee—creating composable workflows without modifying core logic. According to the Yeachan-Heo/oh-my-claudecode source code, each skill operates as an MCP-compatible module that modifies task definitions or wraps results in verification loops.

The Three Layers of Skill Composition

Execution Layer

The execution layer serves as the primary work handler that determines what action to perform. Common execution skills include default for general builds, planner for strategic planning, and orchestrate for coordinating complex multi-step operations. This layer forms the foundation of the skill stack and executes the actual user-requested operation.

Enhancement Layer

Enhancement skills provide optional augmentations that extend or accelerate execution without changing the core task logic. Examples include ultrawork for parallel agent spawning, git-master for atomic commit creation, and frontend-ui-ux for UI scaffolding. These skills modify task definitions—for instance, adding parallel:true parameters—or invoke additional agents during workflow execution.

Guarantee Layer

The guarantee layer acts as an optional safety net that prevents workflow completion until verification succeeds. The ralph skill exemplifies this tier by implementing persistent verification loops that repeatedly check build, test, lint, and functional criteria. According to skills/ralph/SKILL.md【/cache/repos/github.com/Yeachan-Heo/oh-my-claudecode/main/skills/ralph/SKILL.md#L145-L158】, this layer only reports success when all checks pass green.

How Skill Composition Works Internally

The orchestration process follows a strict composition order defined in docs/ARCHITECTURE.md【/cache/repos/github.com/Yeachan-Heo/oh-my-claudecode/main/docs/ARCHITECTURE.md#L81-L102】.

1. Keyword detection parses user prompts to extract skill tokens (e.g., ultrawork:). The detection mechanism resides in src/hooks/AGENTS.md【/cache/repos/github.com/Yeachan-Heo/oh-my-claudecode/main/src/hooks/AGENTS.md#L163-L166】.

2. The system builds a skill stack in the following order:

   [Execution] → [Enhancement-1] → … → [Enhancement-N] → [Guarantee?]

3. Each skill functions as a small MCP-compatible module capable of modifying task definitions, invoking additional agents, or wrapping final results in verification loops.

4. During execution, control flows from the outermost skill inward. The guarantee layer (if present) executes first, followed by enhancements, until the core execution skill finally runs the operation.

5. Post-execution, the guarantee layer runs verification steps. As implemented in ralph, this involves persistent checking of build, test, and lint criteria【/cache/repos/github.com/Yeachan-Heo/oh-my-claudecode/main/skills/ralph/SKILL.md#L145-L158】.

Practical Examples of Skill Composition Layering

Basic Execution-Only Task

For simple operations requiring only core functionality:

/oh-my-claudecode:default "Add a new endpoint to the API"

This invokes only the execution layer (default) without enhancements or guarantees.

Adding Parallelism and Automation

Complex refactoring tasks leverage enhancement layering:

/oh-my-claudecode:ultrawork "refactor the API layer"

This creates a composition where:

• Enhancement: ultrawork spawns multiple agents to refactor files in parallel【/cache/repos/github.com/Yeachan-Heo/oh-my-claudecode/main/skills/ultrawork/SKILL.md#L26-L33】
• Enhancement: git-master automatically creates atomic commits after each sub-task【/cache/repos/github.com/Yeachan-Heo/oh-my-claudecode/main/skills/git-master/SKILL.md#L10-L15】
• Execution: default performs the actual refactoring

Full Stack with Verification Guarantee

Complete workflows utilize all three layers:

/oh-my-claudecode:ralph /oh-my-claudecode:ultrawork "migrate the auth module"

The resulting stack (outermost to innermost):

1. Guarantee: ralph wraps the entire run, entering verification loops post-execution【/cache/repos/github.com/Yeachan-Heo/oh-my-claudecode/main/skills/ralph/SKILL.md#L145-L158】
2. Enhancement: ultrawork executes the migration in parallel across agents
3. Execution: default or orchestrate carries out code changes

The user receives a single command interface while OMC internally constructs the three-layer stack, ensuring the migration reports success only after all build, test, and lint validations pass.

Key Implementation Files

• docs/ARCHITECTURE.md — Defines skill layer hierarchy and composition order【/cache/repos/github.com/Yeachan-Heo/oh-my-claudecode/main/docs/ARCHITECTURE.md#L81-L102】
• src/hooks/AGENTS.md — Implements keyword detection and skill token parsing【/cache/repos/github.com/Yeachan-Heo/oh-my-claudecode/main/src/hooks/AGENTS.md#L163-L166】
• skills/ultrawork/SKILL.md — Parallelism enhancement implementation【/cache/repos/github.com/Yeachan-Heo/oh-my-claudecode/main/skills/ultrawork/SKILL.md#L26-L33】
• skills/git-master/SKILL.md — Atomic commit automation enhancement【/cache/repos/github.com/Yeachan-Heo/oh-my-claudecode/main/skills/git-master/SKILL.md#L10-L15】
• skills/ralph/SKILL.md — Verification guarantee layer with persistence【/cache/repos/github.com/Yeachan-Heo/oh-my-claudecode/main/skills/ralph/SKILL.md#L145-L158】
• src/agents/AGENTS.md — Agent tier mappings for execution skills【/cache/repos/github.com/Yeachan-Heo/oh-my-claudecode/main/src/agents/AGENTS.md#L131-L138】

Summary

• Oh-My-ClaudeCode uses skill composition layering to wrap core execution without switching agents
• The three-tier architecture consists of execution (core work), enhancement (capabilities like parallelism), and guarantee (verification safety nets)
• Skills compose in strict order: execution first, then enhancements, finally optional guarantee
• Each skill operates as an MCP-compatible module that modifies tasks or wraps results
• Keyword detection in src/hooks/AGENTS.md triggers stack construction based on user prompt tokens

Frequently Asked Questions

What is the execution layer in oh-my-claudecode?

The execution layer is the foundational skill tier that handles primary work operations. It includes skills like default for general builds, planner for strategic planning, and orchestrate for complex coordination. This layer determines what action to perform and executes the actual user-requested operation.

How do enhancement skills modify the workflow?

Enhancement skills augment execution by modifying task definitions or invoking additional agents. For example, ultrawork adds parallel:true parameters to enable multi-agent processing, while git-master automatically creates atomic commits after successful sub-tasks. These skills stack between execution and guarantee layers without altering core logic.

Can multiple enhancement skills be used together?

Yes, the architecture supports stacking multiple enhancement skills in sequence. The composition order follows [Execution] → [Enhancement-1] → … → [Enhancement-N] → [Guarantee?], allowing combinations like ultrawork plus git-master to run simultaneously. Each enhancement wraps the previous layer, creating a nested execution flow.

What happens if the guarantee layer verification fails?

When the guarantee layer detects failures, it prevents workflow completion and reports specific verification errors. According to skills/ralph/SKILL.md【/cache/repos/github.com/Yeachan-Heo/oh-my-claudecode/main/skills/ralph/SKILL.md#L145-L158】, the ralph skill repeatedly checks build, test, and lint criteria, persisting in a verification loop until all checks pass green or the user intervenes.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}