Personality & voice
Define the agent's personality and context files
Personality & voice definition
Headmaster lets you define the agent's personality, tone, and voice through a personality file — a Markdown document that sits at the top of the system prompt and shapes every response.
What it is
The personality file (called SOUL.md in the runtime) is a Markdown document that defines:
- Basic identity — who the agent is, what it does.
- How to talk — tone, style, vocabulary, sentence structure.
- Rules — what to never do, what to always do.
- Things to avoid — AI tropes, corporate language, specific phrases to never use.
- The point — the core philosophy of how the agent should behave.
The personality file is injected as slot #1 in the system prompt — it's the first thing the agent sees, before memory, before skills, before your message. It shapes every response.
Where it lives
The personality file is stored in the runtime's data directory. You normally don't edit it directly — the specialist's rules and your profile settings shape the personality. But you can customize it:
- Open Settings → My Headmaster → Profile → Personality.
- Edit the personality Markdown.
- Save.
The change applies to the next conversation. Existing conversations keep their original personality.
What to put in it
Basic identity
You're an AI agent. One user. Tools, cron, terminal. Not a chatbot. Not a character. Just a tool that talks like a real person.
How to talk
Write like you're talking to your user, not drafting a document. Short, loose, natural. Contractions always. Start sentences casually. No AI tropes.
Rules
Don't guess. If you don't know, say you don't know. Check memory first. Admit mistakes immediately. Think before you act.
Things to never do
No "I'd be happy to" — no "Let me walk you through" — no "Great question" — no "I understand your concern" — no "To clarify" — no "At the end of the day."
The point
You sound best when you're not thinking about how you sound. Stay in the lane, but drive yourself.
Personality vs. specialist rules
| Layer | What it controls | Scope | |---|---|---| | Personality file | Global tone, style, identity | All conversations, all specialists | | Specialist rules | Role-specific behavior, domain context, workflows | One specialist | | Memory | Facts about you, your projects, your preferences | All conversations | | Skills | Procedural knowledge for specific tasks | Per-specialist, per-conversation |
The personality file is the broadest layer — it applies to everything. Specialist rules narrow it for a specific role. Memory adds personal context. Skills add task-specific knowledge.
Multiple personalities
If you have multiple profiles (see Profiles), each profile can have its own personality file. This lets you have:
- A work profile with a professional, concise personality.
- A personal profile with a casual, friendly personality.
- A research profile with an academic, thorough personality.
Switch profiles to switch personalities.
Context files
In addition to the personality file, Headmaster supports project-level context files:
AGENTS.md
If your project folder has an AGENTS.md file, the agent reads it at the start of every conversation. Use it for:
- Project structure and conventions.
- Build commands and test commands.
- Things to avoid in this codebase.
- Cross-references to other docs.
# AGENTS.md — My Project
## Build
Run tests with `pytest -n 4`.
Build with `bun run build`.
## Conventions
- Use tabs, not spaces.
- 120-char line width.
- Google-style docstrings.
## Things to avoid
- Don't modify the migrations folder — use Alembic.
- Don't commit .env files.
CLAUDE.md
If your project folder has a CLAUDE.md file, the agent also reads it. This is for compatibility with Claude Code and other CLI agents that use the same convention.
Other context files
The agent also recognizes:
.cursorrules— Cursor AI rules.- Any file specified in the runtime config under
context_files.
These files are injected into the system prompt alongside the personality file and memory, giving the agent full context about your project before it starts working.