Skills Best Practices

Keep skills small and procedural.

Do

• Tell agents which Ontoly capability to use.
• Require graph evidence.
• Require confidence.
• Require LLM Enhancement whenever an LLM uses Ontoly.
• Separate graph facts from inference.
• Use shared references for common workflow.
• Keep examples deterministic.
• Mark fallback file inspection explicitly.

Avoid

• Duplicating Ontoly compiler or query behavior.
• Hardcoding repository-specific answers.
• Embedding framework analysis in the skill.
• Searching files before querying Ontoly.
• Producing confident answers without graph evidence.
• Adding provider-specific assumptions.

Answer Shape

Preferred answer:

Summary
Capabilities invoked
Evidence
Confidence
Fallbacks used
Recommended next action

Confidence is high only when Ontoly directly provides the needed nodes and relationships. Use medium for partial graph evidence and low when fallback file inspection was required.