intersystems-ready-hackathon/ready-2026-team-08/ai-hub-policies

AI Hub Policies

Use this skill for governance over tool execution: allow or deny, sanitize, audit, and compose global or ToolSet-local policies.

When to Use

• Enforcing read-only or restricted-path behavior
• Logging tool calls, errors, or durations
• Attaching global policies to a ToolManager
• Defining ToolSet-local policies in XML

Core Workflow

1. Decide whether the policy is authorization, audit, or discovery-related.
2. Implement the right base class method.
3. Attach the policy globally or within a ToolSet.
4. Validate the policy against the exact tools and arguments it should govern.
5. Compose global and ToolSet-local rules only when the layering is intentional.

Critical Rules

• Use %AI.Policy.Authorization for allow or deny decisions and argument sanitization.
• Use %AI.Policy.Audit for logging and telemetry.
• Use ToolSet-local policies for collection-specific rules and ToolManager policies for global rules.
• Authorization policies run global first and ToolSet second; both must allow.
• Audit policies run at both levels when present.
• ToolSet-local XML-configured policies require the XML-enabled pattern from the advanced guide.

Common Mistakes

• Treating global and ToolSet-local policies as interchangeable
• Forgetting %XML.Adaptor and XML property projection for ToolSet-local policy configuration
• Modifying arguments without validating the downstream tool still accepts the sanitized form
• Adding audit policies without deciding what success, failure, and result size should mean operationally

References

• Policies reference