Agent Best Practices - Noxus Documentation

Follow these best practices to create reliable, efficient, and secure agents in Noxus.

Instructions & Prompts

Practice	What to Do	Why It Matters
Be Specific	Write clear, detailed instructions about agent behavior	Reduces ambiguity and improves consistency
Provide Context	Give agents domain knowledge and background information	Enables more informed and accurate responses
Set Boundaries	Explicitly state what agents should NOT do	Prevents unwanted behavior and maintains control
Define Success	Explain what a successful outcome looks like	Helps agents understand desired results
Use Examples	Show examples of good responses in instructions	Guides agents toward preferred output format

Tool Configuration

Practice	What to Do	Why It Matters
Selective Access	Enable only tools relevant to agent’s purpose	Reduces complexity and potential errors
Permission Controls	Configure appropriate tool permissions	Maintains security and prevents misuse
Usage Limits	Set reasonable limits on tool calls	Controls costs and prevents runaway execution
Monitor Usage	Track which tools are used and how often	Identifies optimization opportunities
Test Tools	Verify tool access works correctly	Ensures agent can complete intended tasks

Testing & Iteration

Practice	What to Do	Why It Matters
Start Simple	Begin with basic configuration, add complexity gradually	Easier to identify what works and what doesn’t
Test Scenarios	Try various conversation paths and edge cases	Uncovers unexpected behaviors
Review Conversations	Regularly check conversation logs	Identifies areas for improvement
Iterate Continuously	Refine instructions based on real usage	Improves quality over time
Monitor Performance	Track response quality and user satisfaction	Measures agent effectiveness

Security & Safety

Practice	What to Do	Why It Matters
Access Control	Implement role-based permissions	Protects sensitive data and operations
Data Handling	Define clear rules for handling user data	Ensures compliance and privacy
Audit Logging	Enable comprehensive logging	Provides accountability and debugging info
Rate Limiting	Set appropriate usage limits per user	Prevents abuse and controls costs
Escalation Rules	Define when to escalate to humans	Ensures critical issues get proper attention
Content Filtering	Implement input/output validation	Prevents inappropriate content

Performance & Cost

Practice	What to Do	Why It Matters
Model Selection	Choose appropriate model for task complexity	Balances quality with cost and speed
Spend Limits	Set a model provider spend limit in Usage controls	Caps model spend for a workspace or the organization
Cache Responses	Cache frequently asked questions	Reduces costs and improves speed
Monitor Costs	Track spending per conversation	Identifies expensive patterns
Optimize Prompts	Keep instructions concise but clear	Reduces token usage without losing quality

Common Pitfalls to Avoid

Instruction Issues

Vague or ambiguous instructions
Instructions that conflict with each other
Missing examples of desired behavior
Not defining agent limitations
Overly complex instructions

Tool Configuration

Enabling too many tools (confuses the agent)
Not testing tool access before deployment
Missing error handling for tool failures
No limits on expensive operations
Tools without clear purpose

Security & Safety

No content filtering or validation
Missing escalation rules
Insufficient access controls
No monitoring or alerting
Ignoring compliance requirements

Next Steps