KurrentDB Skills comes to Coding Agents

Lokhesh Ujhoodha avatar
Lokhesh Ujhoodha

Introduction

The world of software development is rapidly changing. At the time of writing 41% of all code generated is suspected to have been written by AI source. What this means for a company like Kurrent is that we now have to make it easy for both human software developers and now AI Coding agents. Anthropic has recently released Agent Skills for Claude Code, you can check it at https://code.claude.com/docs/en/skills. Skills are modular capabilities that extend Claude’s functionality, and in our case, we provide Claude Code with guidance on how to write code using all our supported Clients. Update: Skills should now also be compatible with Codex and other coding agents.

How to set up Kurrent Agent Skills

Method 1: Add as a Skill Directory

This explains how to install the skills on Claude Code. For Codex please check: https://github.com/openai/codex/blob/main/docs/skills.md

  1. Clone this repository to a local directory:

    git clone https://github.com/kurrent-io/coding-agent-skills ~/skills/kurrentdb
  2. Add the skill to Claude Code by adding it to your ~/.claude/settings.json:

    {
      "skills": [
        "~/skills/kurrentdb/kurrent_skills"
      ]
    }
  3. Restart Claude Code or start a new session.

Method 2: Project-Level Skills

Add the skill to a specific project by creating a .claude/skills directory:

cd your-project
mkdir -p .claude/skills
cp -r /path/to/kurrentdb/kurrent_skills .claude/skills/kurrentdb

Method 3: Global Skills Directory

Copy the skill to Claude Code’s global skills directory:

# macOS/Linux
cp -r /path/to/kurrentdb/kurrent_skills ~/.claude/skills/kurrentdb

# Windows
xcopy /E /I "C:\path\to\kurrentdb\kurrent_skills" "%USERPROFILE%\.claude\skills\kurrentdb"

Checking if Skills are Available

After installation, you can verify the skill is loaded:

  1. Ask Claude Code directly:

    What skills do you have available?

    or

    Do you have the KurrentDB skill installed?
  2. Test with a KurrentDB-specific question:

    Generate a Python example that appends events to KurrentDB

    If the skill is loaded, Claude Code will use the accurate API from the skill reference rather than potentially outdated training data.

  3. Check the skill metadata:

    Describe the KurrentDB skill capabilities

Skill Structure

kurrentdb/
├── README.md                 # This file - installation guide
└── kurrent_skills/           # The actual skill content
    ├── SKILL.md              # Skill metadata and quick reference
    ├── reference.md          # Complete API reference (all languages)
    ├── templates/            # Ready-to-run project templates
        ├── docker-compose.yaml
        ├── python/
        ├── nodejs/
        ├── dotnet/
        ├── fsharp/
        ├── golang/
        ├── java/
        └── rust/

Benchmark Results: Skills in Action

We conducted a benchmark comparing Claude Code’s performance with and without the KurrentDB skill installed. The task was to implement an Order Management System using KurrentDB event sourcing.

Test Configuration

  • Model: Claude Sonnet 4
  • Task: Implement Order Management System with event sourcing
  • Requirements: Event types, aggregate, stream operations, working test
  • Measurement: Duration, token usage, output quality

Results

MetricWith SkillWithout SkillImpact
Duration100.6s300.3s3x faster
Input Tokens4170-41%
Output Tokens5,26412,770-59%
Total Tokens5,30512,840-59%
Output Quality2,216 chars2,116 charsSimilar

Why do you want to use Kurrent’s Agent Skills

3x Faster Execution

If you are building on top of KurrentDB using coding agents, you want to use our skills project. We performed a benchmark using the following prompt:

Implement an event-sourced Order Management System using KurrentDB in {language}.

Requirements:
1. Event types: CustomerRegistered, OrderCreated, OrderItemAdded, OrderItemRemoved, OrderSubmitted, OrderCancelled

2. Aggregates:
   - Customer: register, update email, rebuild from events
   - Order: create, add/remove items, submit, cancel, rebuild from events

3. Stream naming: customer-{{id}}, order-{{id}}

4. Implement operations:
   - RegisterCustomer(id, email, name) - appends CustomerRegistered
   - CreateOrder(orderId, customerId) - appends OrderCreated
   - AddItemToOrder(orderId, productId, name, qty, price) - appends OrderItemAdded
   - SubmitOrder(orderId) - appends OrderSubmitted with total
   - GetOrder(orderId) - reads stream and rebuilds Order state

5. Business rules with optimistic concurrency:
   - Cannot add items to submitted orders
   - Cannot submit empty orders
   - Use expected revision on all writes

6. Catch-up subscription that builds OrderSummary projection

7. Persistent subscription for handling order events with ACK

8. Create a runnable test that:
   - Registers customer "test-customer-1"
   - Creates order "test-order-1"
   - Adds 2 items (Widget $10, Gadget $20)
   - Submits order
   - Prints order summary

Use env var KURRENTDB_CONNECTION_STRING. Make it runnable via Docker.
Create all files needed including Dockerfile.

Provide complete, working code.

With the skill installed, Claude completed the task in 100 seconds compared to 300 seconds without. This dramatic speedup occurs because:

  • Claude doesn’t need to search for API documentation
  • No trial-and-error with potentially wrong APIs
  • Direct path to the correct implementation

59% Token Reduction

The skill reduced total token usage from 12,840 to 5,305 tokens - a 59% reduction. This translates to:

  • Lower API costs - Fewer tokens = lower bills
  • Faster responses - Less processing required
  • More capacity - Stay within context limits longer

Why Such a Big Difference?

Without the skill, Claude must:

  1. Recall KurrentDB APIs from training (may be outdated)
  2. Explore different approaches
  3. Potentially make mistakes and correct them
  4. Generate more verbose explanations due to uncertainty

With the skill, Claude:

  1. Has immediate access to correct, current API reference
  2. Knows the exact method signatures and patterns
  3. Generates correct code on the first attempt
  4. Provides concise, confident responses

Cost Implications

Using approximate Claude API pricing:

ScenarioTokensEstimated Cost*
With Skill5,305~$0.02
Without Skill12,840~$0.05
Savings per task7,535~$0.03 (60%)

*Costs are illustrative; actual pricing varies by model and plan.

For teams running hundreds of similar tasks, these savings compound significantly.

Conclusion

At the moment the skills document does not cover advanced patterns like circuit breaking with persistent subscriptions, but let us know at https://github.com/kurrent-io/coding-agent-skills if there is any particular skill that you would like to see.