Tips & Best Practices

Prompting tips

Be specific

Vague prompts lead to vague results. Instead of:

Fix this code

Try:

Refactor the authenticate function in src/auth.ts to use async/await instead of callbacks. Maintain backward compatibility for the legacy API.

Provide context

Use context files or @-references:

@README.md @package.json what stack is this project using?

Break down complex tasks

For multi-step work, ask Mona to plan first:

Plan the steps to migrate this Express app to Fastify, then execute them one by one.

Use [SILENT] for monitoring

When setting up automated checks:

Check if the site is up. If yes, respond with [SILENT]. If down, describe the error.

CLI shortcuts

Cost optimization

Model selection

Use context efficiently

• Reference specific files instead of entire directories
• Use skills for repetitive workflows
• Clear old sessions to reduce memory overhead

Local models

For cost-sensitive tasks, run a local model:

ollama run llama3:8b
monoclaw config set model.custom.endpoint "http://localhost:11434/v1"

Security best practices

• Use Docker backend for untrusted code
• Enable allowlists on all gateway platforms
• Review before sending — Mona drafts emails but doesn't auto-send
• Rotate master key monthly: monoclaw secrets rotate-master-key
• Audit installed skills — Review ~/.monoclaw/skills/ regularly

Memory management

• Seed USER.md with your preferences at setup
• Use /checkpoint before large refactoring
• Run monoclaw memory tidy weekly
• Export memories before major updates

Gateway setup checklist

• Bot token configured
• Allowlist populated with user IDs
• Admin users designated
• Approval mode set to ask
• Gateway tested in DM before adding to groups
• /update command tested

Troubleshooting workflow

When something breaks:

1. monoclaw doctor
2. Check ~/.monoclaw/logs/
3. Verify config: monoclaw config check
4. Test with minimal setup (no skills, default model)
5. Check provider status page