Module 5: Claude Sub-Agents and Skills

MGMT 675: Generative AI for Finance

Kerry Back, Rice University

Sub-Agents

What Is a Sub-Agent?

When Claude Code handles a complex task, it can spin up sub-agents — separate Claude instances that work on subtasks in parallel or in sequence, then report back.

Orchestrator

• The main Claude session you are talking to
• Breaks the task into pieces
• Delegates to sub-agents and synthesizes results

Sub-Agent

• A fresh Claude instance with its own context
• Receives a specific subtask and tools
• Returns its result to the orchestrator

Sub-agents allow Claude to work in parallel and to keep each subtask focused — the orchestrator stays high-level while sub-agents do the detailed work.

Spontaneous Sub-Agents

Claude Code decides on its own when a task warrants sub-agents. You do not have to ask.

• Parallel research: fetching data from multiple sources simultaneously
• Multi-file analysis: reviewing several documents at once and comparing findings
• Iterative workflows: one agent drafts, another critiques, another revises
• Long computations: splitting a large dataset across agents to stay within context limits

You will see a notice when Claude spawns a sub-agent. Each sub-agent has its own tool permissions and context window, so it starts fresh — it does not inherit your conversation history.

Requesting a Sub-Agent

You can explicitly ask Claude to use a sub-agent for any task:

• “Use a sub-agent to research competitor pricing and report back.”
• “Spawn a sub-agent to review this code for bugs while you work on the next feature.”
• “Use three parallel sub-agents to analyze each of these three reports, then summarize.”

This is especially useful when you want a task isolated from the main context — for example, a long web search or a large file analysis that would otherwise fill the context window.

The /critique skill in the exercise below is a worked example: it spawns three parallel sub-agents (correctness, clarity, devil’s advocate) and has the orchestrator synthesize their findings.

Claude Skills

Claude Desktop vs. Terminal Claude Code

Claude Desktop — Chat tab

• Equivalent to claude.ai in a browser
• Conversational interface, no filesystem access
• Built-in document skills auto-activate
• Custom skills uploaded as ZIP files

Claude Desktop — Code tab

• Equivalent to the Claude Code CLI in a terminal
• Full filesystem and tool access
• Reads ~/.claude/skills/ for custom skills
• Invoke skills with /skill-name

The Code tab and the terminal CLI are the same engine. Everything covered in this module applies equally to both.

Installing Claude Code (Terminal)

Use the native installer — npm is deprecated as of 2026.

macOS / Linux / WSL

curl -fsSL https://claude.ai/install.sh | bash

Windows (PowerShell)

irm https://claude.ai/install.ps1 | iex

Requires Git for Windows installed first.

Then launch from any project folder:

cd my-project
claude

The native installer auto-updates in the background and requires no Node.js. After installing, run claude doctor to verify everything is working.

Built-In Document Skills

Anthropic ships four document skills with every Claude product. No setup required.

Excel

• Create and edit Excel workbooks
• Uses live formulas, not hardcoded values
• Supports charts, formatting, multiple sheets

Word

• Create and edit Word documents
• Professional formatting, headers, tables
• Supports tracked changes and comments

PowerPoint

• Create and edit PowerPoint decks
• Slide layouts, charts, speaker notes
• Can modify existing presentations

Desktop chat / claude.ai

• Skills activate automatically based on your request
• No / command needed

Claude Code (CLI or Desktop Code tab)

• Invoke explicitly: /xlsx, /docx, /pptx
• Or just ask — Claude invokes the right skill

Both surfaces

• Same skill logic and output quality
• Code tab reads ~/.claude/skills/ like the CLI

The /xlsx Skill

In Claude Code, skills are plain text files stored in ~/.claude/skills/. You can read the full /xlsx skill at:

• GitHub: https://github.com/anthropics/skills/blob/main/skills/xlsx/SKILL.md
• In Claude Code on your machine:
  • Windows: C:\Users\yourname\.claude\skills\xlsx\SKILL.md
  • Mac: ~/.claude/skills/xlsx/SKILL.md

It instructs Claude to use live Excel formulas instead of hardcoded values, specifies a standard workflow (create, format, save, recalculate, verify), and includes financial modeling standards.

Claude Desktop and claude.ai use the same skill logic, but the files are managed by Anthropic — you cannot browse them on your machine.

Pre-Installed Skills: Skill Creator

The /skill-creator skill is a guide for building your own skills.

• Walks you through designing a new skill step by step
• Generates the skill definition file
• Installs the skill so it appears as a / command

We will use /skill-creator in the exercise below to build a personal task manager skill.

Custom Skills in Claude Desktop and claude.ai

The built-in document skills are always on. You can also add your own custom skills.

Desktop chat / claude.ai

• Go to Settings → Customize → Skills
• Click + to upload a ZIP file containing a folder with a SKILL.md file
• Toggle custom skills on or off
• Claude auto-invokes based on the skill’s description

Claude Code (CLI or Desktop Code tab)

• Drop a folder with a SKILL.md into ~/.claude/skills/
• Invoke with /skill-name or let Claude load it from context
• Project-level skills go in .claude/skills/ inside your repo

The description in your SKILL.md tells Claude when to invoke the skill, so make it specific. Custom skills are private to your account.

Exercise: A Critique Skill

Create a /critique skill that reviews work Claude has produced by spawning three parallel reviewer agents:

Correctness

• Factual errors, logic gaps
• Missing assumptions
• Math or data mistakes

Clarity

• Structure and readability
• Redundancy or buried points
• Proposed rewrites

Devil’s Advocate

• Challenge assumptions
• Strongest counterarguments
• Weakest reasoning

The skill should synthesize findings, present them ranked by severity, and offer to apply revisions. Ask Claude to create it.

Exercise: Critique Loop

1. Copy a document you’ve created into your working folder (e.g., a memo, report, or exercise solution).
2. Tell Claude: “Apply the critique skill in a loop — critique, make the recommended changes, critique again, and continue until the agents are happy.”

Exercise: Create a Tasks Skill

Use /skill-creator to build a tasks skill that manages personal and school to-do lists. The skill should:

• Store tasks in markdown files (e.g., personal.md, school.md)
• Add new tasks with a category and due date
• List pending tasks when asked
• Mark tasks as complete

Exercise: Steps

1. Describe the task manager you want and ask Claude to create a skill for it
2. Let Claude generate and install the skill
3. Test it: “/tasks Add: finish DCF exercise, due Tuesday”
4. Test it: “/tasks What’s pending?”

This should work fine without typing /skill-creator because Claude knows about the skill-creator skill via the CLAUDE.md.