---

name: blueprints-writing description: Use when creating or updating technical blueprint documentation for new features, API changes, or architectural modifications. Always use search_blueprints first to avoid duplication, then write_blueprint with proper structure. allowed-tools: [Read, Write, Edit, Grep, Glob, mcp__plugin_hashi-blueprints_blueprints__search_blueprints, mcp__plugin_hashi-blueprints_blueprints__read_blueprint, mcp__plugin_hashi-blueprints_blueprints__write_blueprint]

Writing Technical Blueprints

Purpose

Technical blueprints document how systems work internally. They differ from user documentation (how to use) and README files (project overview).

When to Write Blueprints

Write blueprints for:

• Complex systems with multiple components
• Public APIs and their contracts
• Architecture decisions and their rationale
• Behavior that isn't obvious from code
• Integration points between systems

Skip blueprints for:

• Self-documenting code (simple utilities)
• Test files (tests ARE the documentation)
• External dependencies (link to their docs)

Creating Blueprints

IMPORTANT: Always use the write_blueprint MCP tool to create or update blueprints. This automatically adds the required frontmatter.

Copy// Before writing, search for existing blueprints
const existing = await search_blueprints({ keyword: "auth" });

// Read existing blueprint if updating
const current = await read_blueprint({ name: "authentication" });

// Write or update the blueprint
await write_blueprint({
  name: "system-name",
  summary: "Brief one-line description",
  content: "# System Name\n\n## Overview\n..."
});

The MCP tool handles frontmatter automatically - you just provide the content.

Blueprint File Structure

When writing blueprint content (passed to write_blueprint), use this structure:

Copy# System Name

One-line description of what this system does.

## Overview

2-3 paragraphs covering:
- Why this system exists (problem it solves)
- What it does at a high level
- Where it fits in the larger architecture

## Architecture

### Components

List major components:
- **ComponentA** - Purpose and responsibility
- **ComponentB** - Purpose and responsibility

### Data Flow

Describe how data moves through the system:
1. Input arrives via X
2. ComponentA processes and transforms
3. Result passed to ComponentB
4. Output returned/stored

### Dependencies

- **Dependency** - Why it's needed, what it provides

## API / Interface

### Functions

#### `functionName(param1, param2)`

Clear description of what the function does.

**Parameters:**
- `param1` (Type) - What this parameter controls
- `param2` (Type, optional) - Default behavior if omitted

**Returns:** Type - Description of return value

**Throws:** ErrorType - When this error occurs

**Example:**
```typescript
const result = functionName('value', { option: true });

Configuration

Behavior

Normal Operation

Step-by-step description of typical execution flow.

Error Handling

How the system responds to:

• Invalid input
• Missing dependencies
• External failures

Edge Cases

Document non-obvious behaviors:

• Empty input handling
• Concurrent access
• Resource limits

Files

Key files and their roles:

• src/main.ts:1-50 - Entry point, initialization
• src/processor.ts - Core processing logic
• src/types.ts - Type definitions

Related Systems

Example format:

Copy- [Related System](./related-system.md) - How they interact

Copy
## Writing Style

### Be Precise

Bad: "The system handles errors gracefully"
Good: "Invalid input returns a ValidationError with the field name and reason"

### Document Behavior, Not Implementation

Bad: "Uses a for loop to iterate through items"
Good: "Processes items sequentially, stopping at the first failure"

### Include Examples

Show, don't just tell:
```typescript
// Good: concrete example
const result = processItems(['a', 'b', 'c'], { parallel: true });
// Returns: { processed: 3, failed: 0 }

Keep Current

• Update blueprints alongside code changes
• Remove documentation for deleted features
• Mark deprecated features clearly

Common Mistakes

1. Too much detail - Don't document every line of code
2. Too little context - Explain why, not just what
3. Stale documentation - Outdated docs are worse than none
4. Duplicate content - Link instead of copying
5. Implementation focus - Document contracts, not internals