Prompts and project rules

Add Veracity-specific conventions to your AI tool by placing a prompt file in your repository, so every session starts with the context your agent needs to generate correct Veracity integration code.

A prompt is reusable instruction text you keep in your repository and reference from your AI tool — a Cursor project rule, a GitHub Copilot instructions file, a Claude Code configuration file, or the equivalent for any other supported agent. Unlike a skill, which an agent loads on demand for a specific task, a project-level prompt applies to every interaction in the repository. Use prompts to establish baseline Veracity conventions: correct terminology, authentication patterns, tenancy requirements, and credential handling rules.

How to add a prompt to your project

Place the prompt file at the path your agent reads. The path differs per tool:

Tool	File path	Scope
GitHub Copilot	.github/copilot-instructions.md	All Copilot interactions in the repository
Cursor	.cursor/rules/veracity.md	All Cursor interactions in the workspace
Claude Code	CLAUDE.md	All Claude Code interactions in the repository
Codex	AGENTS.md	All Codex interactions in the repository
OpenCode	AGENTS.md	All OpenCode interactions in the repository
Windsurf	.windsurfrules	All Cascade interactions in the workspace

Create the file if it does not exist. If the file already exists, append the Veracity prompt content to it.

Baseline prompt: Veracity conventions

Copy this into your project prompt file. It establishes the conventions that apply to all Veracity development tasks in the repository.

## Veracity development conventions

When generating code for this project:

- This project integrates with Veracity by DNV. Follow current official Veracity documentation.
- Use the Veracity Development MCP Server to retrieve current endpoint and authentication documentation before writing any API call.
- Never inline credential values. Read all Veracity client IDs, client secrets, and API keys from environment variables.
- Pass tenant identifiers as API parameters on every Veracity API call. Do not rely on implicit defaults.
- Use the authorisation code flow for user-facing authentication. Use the client credentials flow for service-to-service calls.
- Use the most restrictive workspace access level for each service account: Reader access to specific datasets where possible, workspace-wide access only when required.
- Handle HTTP 401 (invalid or expired token) and HTTP 403 (insufficient permission) as distinct error cases.

Task-specific prompts

For prompts you paste into your agent's chat when working on a specific task, see Common AI tasks. That page groups prompts by domain (identity, access management, Data Workbench) and includes verification checklists for each.

Limitations

• Project-level prompt files apply to all interactions in the repository. If your repository contains non-Veracity code alongside Veracity integrations, the prompt may affect unrelated code generation. Scope with an applyTo pattern if your tool supports it.
• Prompts instruct the agent but do not guarantee the output is correct. Always verify generated code against current Veracity documentation: see Verify generated code.
• Some agents merge the project instructions file into the model's context window on every interaction. Long prompt files consume tokens on each request. Keep the baseline prompt concise and use task-specific inline prompts for detail.