Best Practices

Getting Started Right

1. Initialize every project: helixmind init for project-local memory
2. Write good context: .helixmind/context.md with tech stack, architecture, conventions
3. Feed key files early: helixmind feed README.md --deep src/core/

Daily Workflow

When	What
Start of day	hx: the AI knows your project from yesterday
During development	Let AI explain code, write tests first, check /spiral
End of day	Run /security in the background

Spiral Memory Tips

• Let it learn naturally: Don't front-load everything at once
• Use semantic search: helixmind spiral search "how we handle auth"
• Check periodically: /spiral shows what the AI knows

Agent Interaction

• Be specific: Instead of "Fix the bug" say "Fix the auth middleware. JWT tokens expire too early"
• Let it explore first: "Read the auth module and explain the token refresh"
• Skip-permissions for flow: /skip-permissions when actively watching
• Review before committing: /diff, Validation Matrix, run tests

Choosing the Right Model

Task	Recommended
Quick fixes	Haiku / GPT-4o-mini
General coding	Sonnet / GPT-4o
Complex architecture	Opus
Offline / privacy	Ollama (qwen2.5-coder:32b)

Switch models with /model.

Team Workflows

• Shared context: Commit .helixmind/context.md for same project instructions
• Knowledge transfer: Senior exports, new member imports
• Code review with AI: hx -m "Review the changes in the last 3 commits"

Security

1. Run /security regularly (at least weekly)
2. Never commit API keys. use environment variables
3. Start with safe permissions, only escalate when needed
4. Review YOLO mode output carefully
5. Use checkpoint system. revert unwanted changes immediately