Fix an AI tool call with missing or unsafe arguments

An assistant proposes creating a story with an invalid category or requests access to a record belonging to another account. A well-formed tool call is not sufficient authorisation to execute it.

Why this happens

Models can supply incomplete, incorrect or contextually inappropriate arguments. Tool descriptions guide generation but do not enforce business rules or access control.

Step-by-step solution

1. Define a narrow argument schema with required fields and enumerated choices where appropriate. Avoid a generic execute-anything tool.

2. Validate arguments on the server and resolve ownership using the authenticated session. Do not trust a user ID supplied by the model as proof of permission.

3. Return a structured validation error that supports correction. Keep irreversible actions behind the workflow’s approval step and use idempotency for retries.

Worked example

A publish_story tool should accept a story identifier, then check that the caller can publish that story and that its current state permits publication. The model should not be able to grant itself an admin role.

Verify the fix

Test missing IDs, another user’s ID, invalid status transitions and repeated calls. Each case should produce a predictable outcome without unauthorised writes.

Prevent a repeat

Version tool schemas and maintain permission tests independently of prompt tests. Restrict tool results to the data the assistant actually needs.

Reference

Official documentation or standards guidance

An original Gistmax troubleshooting guide. Scenarios and sample values are illustrative; configuration depends on your software version and environment. Cover diagram: Gistmax Tutorials.