Agent Tool Creation

Create sandboxed tools that agents can call with explicit inputs, permissions, verification, and testing.

​

Open the tool builder

1. Open Organization → Agent Tools → Create Tool, or visit /my-org/tools/new.
2. Enter a descriptive Tool Name, choose a Category, and describe the purpose.
3. Add tags for discovery and review.

​

Define inputs and permissions

1. Add each required value under Input Parameters.
2. Give each input a stable name, type, and useful description.
3. Choose the least-privileged Permission Level (Sandbox Tier).
4. Start with Restricted (Tier 0) for pure computation with no I/O or network access.
5. Request a higher tier only when the use case and approval process require it.

​

Implement and verify the tool

1. Write or generate the implementation in Tool Code.
2. Validate inputs before use and return a predictable result shape.
3. Select Verify to analyze the tool.
4. Use the test runner with representative JSON input.
5. Fix warnings and retest until behavior is understood.
6. Select Save only when the code, inputs, permissions, and verification results are ready for review.

​

Tool review checklist

• Does the tool do one clearly defined job?
• Are inputs validated and documented?
• Does the code avoid unnecessary I/O, network, or external side effects?
• Is the sandbox tier the minimum required?
• Have normal, invalid, and boundary-case inputs been tested?
• Is the tool ready for organization review and approval?