Claude Code Skills

Kitaru provides two Claude Code skills that teach Claude how to design and build durable agent workflows.

Available skills

Scoping skill

The kitaru-scoping skill helps you decide whether your agent workflow benefits from durable execution, then designs the flow architecture — checkpoint boundaries, wait points, artifact strategy, and MVP scope.

It runs a structured interview and produces a flow_architecture.md specification that feeds directly into the authoring skill.

Use it when:

• You're starting a new agent project and aren't sure where durability adds value
• You have a complex workflow and need help identifying checkpoint/wait boundaries
• You want to validate that Kitaru is the right tool before writing code

Authoring skill

The kitaru-authoring skill teaches Claude how to write valid Kitaru flows, checkpoints, waits, and adapter usage patterns.

What it covers:

• @flow and @checkpoint roles
• kitaru.wait() flow-level guardrails
• kitaru.log(), kitaru.save(), kitaru.load(), kitaru.llm()
• .submit() / .result() checkpoint concurrency
• PydanticAI adapter guidance (kitaru.adapters.pydantic_ai.wrap(...))
• Common mistakes (nested flows, non-serializable checkpoint outputs, etc.)

Recommended workflow

1. Scope first — use /kitaru-scoping (or let Claude trigger it automatically) to validate fit and define your flow architecture
2. Author second — use /kitaru-authoring to build the flows defined in your flow_architecture.md

Install from the marketplace

The Kitaru skills are published as a Claude Code plugin. Add the marketplace and install with two commands:

/plugin marketplace add zenml-io/kitaru-skills
/plugin install kitaru@kitaru

Once installed, Claude Code will automatically use the skills based on context. You can also invoke them explicitly with /kitaru-scoping or /kitaru-authoring.

Manual installation

If you prefer to copy the skill files directly:

mkdir -p .claude/skills/kitaru-scoping .claude/skills/kitaru-authoring

curl -fsSL https://raw.githubusercontent.com/zenml-io/kitaru-skills/main/skills/kitaru-scoping/SKILL.md \
  -o .claude/skills/kitaru-scoping/SKILL.md

curl -fsSL https://raw.githubusercontent.com/zenml-io/kitaru-skills/main/skills/kitaru-authoring/SKILL.md \
  -o .claude/skills/kitaru-authoring/SKILL.md

Good prompts to pair with the skills

Scoping:

• "I want to build a research agent — is Kitaru right for this?"
• "Help me figure out where to put checkpoints in my coding agent workflow."
• "I have a complex agent with 10 steps — help me scope it down."

Authoring:

• "Refactor this script into one @flow with explicit checkpoints."
• "Add a flow-level kitaru.wait() approval gate before publish."
• "Wrap this PydanticAI call in a @checkpoint and preserve replay safety."

Related pages

• MCP Server
• Execution management