Hermes Agent Tips & Best Practices Summary

Disclaimer: Source linked at bottom. Illustrations created by me.

Getting Best Results

• Be specific about desired outcomes with context (file paths, errors)
• Use AGENTS.md for recurring project instructions
• Let agent utilize its tools and skills for complex tasks
• Provide context upfront to reduce iteration cycles

CLI Power User Tips

• Multi-line input: Alt+Enter, Ctrl+J, or Shift+Enter
• Paste detection: pastes buffered as single message
• Interrupt with Ctrl+C (double for force exit)
• Resume sessions with hermes -c or hermes -r “title”
• Clipboard image paste via Ctrl+V
• Slash command autocomplete with Tab

Context Files

• AGENTS.md: project-specific instructions (auto-loaded)
• SOUL.md: global personality customization
• .cursorrules compatibility
• Lazy loading of subdirectory AGENTS.md during tool calls

Memory & Skills

• Memory: facts/preferences (~2,200 chars)
• Skills: procedures/workflows
• Create skills for repeated 5+ step tasks
• Manage memory via “clean up memory” commands
• Agent remembers key takeaways when prompted

Performance & Cost

• Stable system prompt enables caching
• Use /compress to reduce token count
• Delegate tasks for parallel work
• Use execute_code for batch operations
• Select appropriate model complexity per task

Messaging Tips

• Set home channel for proactive outputs
• Use /title to organize sessions
• DM pairing for secure team access
• Control tool visibility with /verbose

Security

• Use Docker sandbox for untrusted code
• Windows: enforce UTF-8 encoding
• Review before “always” approving dangerous commands
• Command approval safety net (skip in containers)
• Use platform-specific allowlists for bots

Source > https://hermes-agent.nousresearch.com/docs/guides/tips