Skip to content

System Prompts Architecture

Overview

InfluencerPy uses a structured system prompt architecture that separates concerns into four distinct components. This ensures users can customize their scout's behavior without accidentally breaking system guardrails or platform formatting rules.

Prompt Components

1. General Instructions (Hidden)

Purpose: Core system guardrails and professional behavioral guidelines.

Content: - Be objective and fact-based - Prioritize quality over quantity - Respect copyright and attribution - Avoid clickbait or sensationalism

Visibility: Hidden from users, automatically included in all scouts.

2. Tool Instructions (Hidden)

Purpose: Auto-generated guidance for using enabled tools.

Content: Varies based on scout type and tools: - google_search: How to formulate search queries - rss: How to parse feed items - reddit: How to browse subreddit discussions - arxiv: How to search academic papers - browser [EXPERIMENTAL]: How to navigate and extract content

Visibility: Hidden from users, automatically generated based on configured tools.

3. Platform Instructions (Hidden)

Purpose: Platform-specific formatting and constraints.

Supported Platforms:

X (Twitter): - Maximum 280 characters per post - 2-3 hashtags maximum - Emojis for engagement - Threading for longer content

LinkedIn: - Professional tone - Up to 3000 characters - Focus on insights and takeaways - Strategic use of line breaks

Visibility: Hidden from users, automatically applied based on target platform.

4. User Instructions (Visible & Editable)

Purpose: Your custom scout goal.

Examples: - "Find breaking AI safety research" - "Summarize this content for a technical audience" - "Find controversial blockchain discussions"

Visibility: This is the ONLY component users see and edit in the UI.

Label in UI: "User Instructions (Your Scout Goal)"

How It Works

Prompt Construction

When a scout runs, the system builds the final prompt using the SystemPrompt dataclass:

from influencerpy.core.prompts import SystemPrompt
from influencerpy.types.prompts import (
    GENERAL_GUARDRAILS,
    build_tool_prompt,
    get_platform_instructions
)

system_prompt = SystemPrompt(
    general_instructions=GENERAL_GUARDRAILS,
    tool_instructions=build_tool_prompt(["rss", "google_search"]),
    platform_instructions=get_platform_instructions("x"),
    user_instructions="Find trending AI news"
)

final_prompt = system_prompt.build(
    date="2025-11-28",
    limit=10
)

Example Output

The final prompt sent to the AI looks like:

You are a professional content scout and curator for social media.

CORE PRINCIPLES:
- Be objective and fact-based in your analysis
- Prioritize quality over quantity
- Respect copyright and provide proper attribution
- Avoid clickbait or sensationalism

AVAILABLE TOOLS:

TOOL: rss
Use this to fetch feed items from the configured RSS source.

TOOL: google_search
Use this to find recent news, articles, and web content.

YOUR GOAL: Find trending AI news

OUTPUT FORMAT FOR X (TWITTER):
- Maximum 280 characters per post
- Use 2-3 relevant hashtags maximum
- Include emojis strategically for engagement

CONTEXT:
date: 2025-11-28
limit: 10

Benefits

For Users

✅ Simple UX: Only edit your custom goal, not system internals
✅ Protected: Can't accidentally break guardrails or formatting
✅ Platform-Aware: Posts automatically formatted correctly
✅ Tool-Aware: Instructions auto-update when tools change

For Developers

✅ Maintainable: All prompt logic centralized
✅ Testable: Each component independently verifiable
✅ Extensible: Easy to add new platforms or tools
✅ Type-Safe: Dataclass ensures valid composition

Customization

Editing User Instructions

  1. In the CLI, select Update Scout
  2. Choose User Instructions
  3. Enter your custom goal

Your instruction will be combined with system guardrails, tool guidance, and platform formatting automatically.

Adding New Platforms

To add a new platform (e.g., Instagram):

  1. Edit src/influencerpy.types.prompts.py

  2. Add to PLATFORM_INSTRUCTIONS dictionary: 

    "instagram": """OUTPUT FORMAT FOR INSTAGRAM:
- Visually-driven content
- Up to 2200 characters
- Use emojis liberally
- Hashtags at the end (up to 30)"""

  3. No other code changes needed!

Migration

Existing scouts are fully backward compatible. The prompt_template field is now interpreted as "User Instructions" and combined with system components automatically.