Build with AI (Skill, Prompting)

You can use AI coding assistants to generate, debug, and explore CRE workflows. There are two approaches: install a skill in your code editor agent or load CRE context into any LLM chat and prompt from there.

Code editor skill

Agent skills are structured instruction sets that give an AI coding agent domain-specific knowledge and a defined approach for a task. When you install a skill, your agent reads it at the start of a session and uses it to guide how it fetches docs, generates code, and debugs — without you having to repeat that context every time. See the Agent Skills specification for the full format and best practices.

Supported agent tools

The skill works with any agent-capable AI coding tool that implements the Agent Skills specification. Verified with Claude Code, Cursor, Codex, Gemini, Copilot, and Cline.

Install the skill

Run the following command to add the CRE skill to your code editor agent:

npx skills add https://github.com/smartcontractkit/chainlink-agent-skills --skill chainlink-cre-skill

Prompt your agent

Once installed, your agent will automatically discover and activate the skill based on your task. For best results across tools, explicitly reference it in your prompt:

Use the /chainlink-cre-skill skill and [describe what you want to build]

In Claude Code, you can invoke it directly as a slash command: /chainlink-cre-skill.

What the skill covers

The skill is organized into focused reference areas, each covering a specific part of CRE. When you ask a question, your agent matches it to the most relevant area and fetches only what it needs — so you get a targeted answer without the agent reading the entire SDK. The areas the skill covers:

Workflow generation

When you ask the skill to build a workflow, it will:

1. Confirm whether you want Go or TypeScript
2. Ask whether HTTP requests should use regular or Confidential HTTP (for privacy-preserving API calls via enclave execution)
3. Generate the complete workflow scaffold immediately — no speculative fetching
4. Give you a cre workflow simulate command to test it
5. Iterate on errors by fetching exactly what the error names, fixing it, and asking you to re-run

Debugging

When diagnosing errors, the skill identifies the capability involved, fetches the relevant doc page, checks repo examples if needed, and proposes a fix based on CRE-specific constraints — not general programming assumptions.

Prompting

If you prefer to work directly in an AI assistant, you can load the full CRE Go documentation into the context window and prompt from there. This is useful for broad exploration, but loads the entire SDK reference upfront — the skill approach above keeps your context window focused on your current task.

Context file: https://docs.chain.link/cre/llms-full-go.txt

Alternatively, you can click the "Copy Page" button in the top right corner of any Chainlink docs page to load the full page content into the context window of your preferred AI assistant. Loading a single page — for example, just the triggers reference or the EVM client page — rather than the entire SDK consumes fewer tokens (lower cost per request for pay-per-token APIs), keeps you within the limits of smaller context windows, and gives the model more relevant signal to attend to, which tends to produce more accurate output.

Copy the starter prompt below, replace the placeholder at the end with your use case, and paste it into your preferred AI assistant.

I'm building a Chainlink CRE workflow using the Go SDK.

Here is the full CRE Go documentation:
https://docs.chain.link/cre/llms-full-go.txt

Please fetch and read that file. After you've read it, please:
- Help me build a CRE workflow based on what I describe
- Answer any questions I have about the CRE Go SDK
- Guide me through implementation step by step

I want to build: [describe your use case]