Agent Ready

AI agents are good at running commands. They are less good at knowing which commands to run, in what order, and what to do when something unexpected happens.

FineType ships with Claude Code skills — structured workflow guidance that teaches agents the complete pipeline from raw CSV to typed DuckDB table. Agents that use FineType with these skills don't stop at profile — they complete the full journey.

What Are Skills?

Skills are markdown files that live alongside FineType's source code. When an agent works in a project with FineType installed, Claude Code automatically discovers the skills and uses them to make better decisions.

No MCP server. No API keys. No configuration. Just a CLI and a skill file.

Available Skills

/finetype-pipeline — The Full Pipeline

Guides agents through every step of the FineType workflow:

profile → schema → validate → load

The skill teaches agents:

• When to investigate — confidence below 90% is a data quality signal, not a FineType error
• What to do with invalid rows — load the valid rows, route invalids to a review queue
• How to interpret the schema — x-finetype-label, x-finetype-transform, and what required means
• When to skip the quality gate — exploratory work can go straight from load, but production data needs validate

/finetype-cli — CLI Reference

A complete reference of every FineType command, flag, and output format. Agents consult this when they need to look up a specific option — like --delimiter for TSV files or --enum-threshold for low-cardinality columns.

Getting the Skills

The skills ship with FineType's source repository. Clone the repo and the skills are ready:

git clone https://github.com/meridian-online/finetype.git

The skills live at:

finetype/
  .claude/
    skills/
      finetype-pipeline/SKILL.md
      finetype-cli/SKILL.md

Claude Code discovers them automatically when you open a session in the repository or any project that includes FineType as a dependency.

Why CLI, Not MCP?

FineType has an MCP server (finetype mcp), and it works well. But skills + CLI offer something MCP cannot:

• Zero setup — no server to start, no transport to configure, no ports to manage
• Universal — any agent with shell access can use the CLI, not just MCP-compatible clients
• Inspectable — the skill files are plain markdown, readable by humans and agents alike
• Composable — pipe FineType output to jq, duckdb, or any other CLI tool

MCP is still available for agents that prefer it. Skills are the recommended path for most workflows.

The Pipeline in Action

Here is what an agent does when it follows the pipeline skill:

# Step 1: Understand the data
finetype profile -f sales.csv

# Step 2: Capture the contract
finetype schema sales.csv

# Step 3: Quality gate — split clean from dirty
finetype validate sales.csv sales.schema.json

# Step 4: Load the clean rows into DuckDB
finetype load -f sales.valid.csv > load.sql
duckdb analytics.db < load.sql

Four commands. Every column typed. Every bad row caught. Every cast guaranteed to succeed.

Without the skill, most agents run finetype profile and stop — leaving the data untyped and unvalidated. The skill ensures agents complete the work.

What's Next