Blitz

System Prompts

Legion CLI assembles system prompts from multiple sources to provide comprehensive context to the AI model.

Prompt Assembly Order

The system prompt is built in src/session/system.ts and src/session/llm/request.ts:

┌─────────────────────────────────────┐
│ 1. Soul (identity & personality)    │  src/kilocode/soul.txt
├─────────────────────────────────────┤
│ 2. Brain (behavioral rules)         │  src/kilocode/brain.txt
├─────────────────────────────────────┤
│ 3. Provider prompt                  │  src/session/prompt/*.txt
├─────────────────────────────────────┤
│ 4. Memory (LEGION.md + session)     │  src/kilocode/memory/
├─────────────────────────────────────┤
│ 5. Environment (context info)       │  src/kilocode/system-prompt.ts
├─────────────────────────────────────┤
│ 6. Skills (available skills)        │  src/skill/
├─────────────────────────────────────┤
│ 7. ELC Rules (language-specific)    │  src/kilocode/elc/rules.ts
├─────────────────────────────────────┤
│ 8. Context Rules (lazy senior dev)  │  src/kilocode/context/rules.ts
├─────────────────────────────────────┤
│ 9. User/system instructions         │  From config and user input
└─────────────────────────────────────┘

Prompt Sources

Soul (`src/kilocode/soul.txt`)

Core identity and personality. Defines who the AI is and its fundamental behavior.

Brain (`src/kilocode/brain.txt`)

Behavioral rules and safety guidelines. Controls how the AI operates, what it can/cannot do, and safety constraints.

Provider Prompts (`src/session/prompt/*.txt`)

Model-specific prompts optimized for different providers:

File	Target
`anthropic.txt`	Anthropic Claude models
`gpt.txt`	OpenAI GPT models
`gemini.txt`	Google Gemini models
`codex.txt`	Codex models
`beast.txt`	High-capability models
`ling.txt`	Ling models
`kimi.txt`	Kimi models
`kilocode-gpt-5.5.txt`	GPT-5.5 specific
`copilot-gpt-5.txt`	Copilot GPT-5
`trinity.txt`	Trinity models
`default.txt`	Fallback default

Memory (`src/kilocode/memory/`)

Project-specific context loaded from:

LEGION.md in project root
~/.legion/LEGION.md (global)
Session memory entries from ~/.legion/legion-memory.json

Environment (`src/kilocode/system-prompt.ts`)

Runtime context injected automatically:

Model information
Platform details
Git status
Current date
Project config paths

ELC Rules (`src/kilocode/elc/rules.ts`)

Language-specific coding rules from Everything Legion Code:

Common rules (always loaded)
Language-specific rules (loaded based on file extensions)

Context Rules (`src/kilocode/context/rules.ts`)

Lazy senior dev mode rules for efficient coding patterns.

Prompt Types

Soul Prompt

You are Legion, a highly skilled software engineer...

Brain Prompt

Contains behavioral rules, safety guidelines, tool usage policies, and coding conventions.

Provider Prompt

Model-specific instructions optimized for each AI provider’s strengths and limitations.

Memory Prompt

# Project Memory
The following memory has been loaded from LEGION.md and session memory.
Use this context to inform your responses about project conventions...

# LEGION.md content
...

Customization

Project-Level

Create LEGION.md in project root:

# Project Conventions
- Use TypeScript strict mode
- Follow Airbnb style guide
- All functions must have JSDoc comments

Global-Level

Create ~/.legion/LEGION.md:

# Global Preferences
- Prefer functional programming
- Use async/await over callbacks
- Always handle errors explicitly

Provider-Specific

The system automatically selects the appropriate provider prompt based on the model being used.

This site is open source. Improve this page.