Back to Blog
ContinuePromptsDeep Dive

Continue Dev Prompts: A Technical Deep Dive

By PRPM Team•October 26, 2025•10 min read

Continue is a VS Code extension that brings ChatGPT-style AI assistance directly into your editor. Unlike other AI coding tools, Continue emphasizes customizable prompts and context providers to give developers full control over AI interactions.

Introduction

Continue's prompt system is slash command-based, context-aware, developer-friendly, and composable. The philosophy: give developers building blocks to create their own AI workflows.

Format Specification

Continue has two types of content: prompts (slash commands) in .continue/prompts/ and rules in .continue/rules/. Both use YAML frontmatter followed by markdown content, but with different metadata requirements.

Prompts (Slash Commands)

Prompts require YAML frontmatter with name, description, and invokable: true:

---
name: Explain Code
description: Explains the selected code in detail
invokable: true
---
Please explain the following code:
{{selectedCode}}

Rules

Rules require YAML frontmatter with name, globs (file patterns), alwaysApply, and description:

---
name: TypeScript Best Practices
globs:
- "**/*.ts"
- "**/*.tsx"
alwaysApply: false
description: TypeScript coding standards
---
Always use strict type checking.
Avoid using 'any' type.

Template Variables

Continue supports template variables for dynamic content:

{{selectedCode}} - Currently selected code
{{currentFile}} - Full content of current file
{{currentFileName}} - Name of current file
{{currentFilePath}} - Path of current file
{{clipboardContent}} - Content from clipboard
{{userInput}} - User's input after slash command

Prompts vs Context Providers

Prompts are pre-defined AI instructions invoked via slash commands (like /explain-code, /write-tests).

Context Providers are dynamic data sources attached to prompts (like @docs, @codebase, @git).

Combining them:

/explain-code @docs @codebase

This runs the "explain-code" prompt with relevant documentation and similar code from the codebase.

PRPM's Implementation

PRPM implements full Continue support with specialized converters that handle the distinctions between prompts and rules:

Conversion Logic

When converting to Continue format, PRPM automatically generates the correct YAML frontmatter based on package subtype:

// Prompts (slash-command or prompt subtype)
if (pkg.subtype === 'slash-command' || pkg.subtype === 'prompt') {
// Generate YAML with: name, description, invokable: true
}
// Rules (rule subtype)
if (pkg.subtype === 'rule') {
// Generate YAML with: name, globs, alwaysApply, description
}

Metadata Mapping

For rules, PRPM extracts glob patterns and alwaysApply flags from the canonical package metadata:

const globs = pkg.metadata?.globs || ['**/*'];
const alwaysApply = pkg.metadata?.alwaysApply ?? false;

Template Variable Preservation

Template variables are preserved as plain text during conversion - no parsing or substitution. They survive roundtrip conversion perfectly.

Best Practices

Use Descriptive Slash Command Names

❌ Bad: Generic names

/explain, /help, /do

✅ Good: Specific, actionable names

/explain-code, /generate-tests, /review-pr, /refactor-function

Set Appropriate Temperature

Match temperature to task: 0.0-0.3 for deterministic tasks (explanation, analysis), 0.4-0.6 for balanced tasks (refactoring), 0.7-1.0 for creative tasks (code generation, architecture design).

Conclusion

Continue's prompt system represents a developer-first approach with slash commands, template variables, markdown format, and context providers. PRPM's implementation leverages format similarity through aliasing to the Claude parser with zero duplication.

Continue Exploring

Windsurf Deep DiveView All Posts