Skip to content

Blog

Slash commands, MCP, sub-agents and skills form a hierarchy

Teams build the same workflow three ways because they treat Claude Code's four overlapping features as equals. They form layers: prompts at the base, MCP and sub-agents beside each other, skills composing the rest.

Written by

Coding agent setup

Published: 4 December 2025 Updated: 2 min read

Engineering leaders keep asking me: “Should we build this as a skill, an MCP server, or a sub-agent?”

Your team has probably already built the same workflow three different ways. The right choice is unclear because Claude Code has four overlapping features, and your engineers are drowning in options.

Treating the features as equals duplicates work

The real problem is that the hierarchy is not understood. When you treat all features as equals, git worktree handling gets built as a skill, a sub-agent and a slash command, and the team wastes time building the same thing three ways.

Most teams jump straight to skills without understanding that they are a composition layer, not a replacement. There’s a hierarchy. Understand it, and the confusion goes away.

The composition hierarchy

Layer 1: slash commands (prompts)

The foundation. A slash command is a prompt the agent executes when triggered, for one-off tasks: commit messages, components, summaries. The prompt is the fundamental unit of knowledge work. Not skills, not MCP: the prompt. Nail this first.

Layer 2A: MCP servers

For external integration: Jira, databases, APIs, third-party services. Integration, not orchestration.

Layer 2B: sub-agents

For parallel work in isolation: fixing 50 tests at the same time, security audits across systems. The context is clean, but it is lost after completion.

Layer 3: skills

For complex, repeating problem sets that need multiple operations. Git worktrees are an example: create, list, remove, merge, manage environment variables. That’s a problem set. One prompt can’t handle it. Skills orchestrate multiple capabilities.

The decision tree

  • External integration: MCP server
  • Parallel isolation: sub-agent
  • One-off task: slash command
  • Repeating problem set: skill, which composes the others

Skills compose slash commands, they don’t replace them

The critical mistake: most teams replace slash commands with skills. Skills compose slash commands.

The right way: /create-worktree and /list-worktrees exist, and a skill wraps both. The wrong way: a skill does everything, with no building blocks, as a monolith. One is modular. The other is a maze.

Start with the primitive

Understand the four layers. Nail the primitive, the prompt. Compose thoughtfully. Start simple: build slash commands first, and move to skills only when you manage repeating problem sets.

Teams with clear hierarchies ship faster. Everyone knows which feature fits the job, and nothing gets built twice.

Are your engineers building the same capability three different ways, or do you have a clear hierarchy for when to use each feature?

Further reading

  1. For tools you own, build a CLI before an MCP server Every MCP server loads all its tools into context before the agent starts. Keep MCP for external services and give your own tools a CLI, scripts or skills, which cost a fraction of the context. 2 min readShort post
  2. Shallow agents fail at step 20, deep agents plan and remember The prompt, tool, result loop works for 5 to 15 steps and collapses on real work. Deep agents add explicit planning, delegation to sub-agents, persistent memory and heavy context engineering. 2 min readShort post
  3. One general agent with a skill library beats 47 custom agents Teams cram every procedure into monolithic agent prompts. Agent Skills replace them with one general agent and a composable library that loads knowledge in layers. 2 min readShort post

All 14 posts on Coding agent setup

ClientsClient work

Scaling needs named owners

Coding agents scale across teams when named people own the platform, the team practice, the guardrails, and the measurement. Rollout & scale is where that ownership is built and handed to internal owners, in your stack.

Training on this topic: Agentic Engineering for Developers and QA
See Rollout & scale

Want to check one team first? The self-check: twelve yes or no questions, three minutes, instant result, no account. Take the self-check