I connected my audit Actor to Claude, and it audited three packages nobody asked for

Description and use

Fixing the input schema in an Apify Actor connected to Claude stopped unintended package audits, aligning results with caller intent.

Audit Actors were unintentionally exposing additional package data due to an input schema misconfiguration, leading to unexpected audit results.

Correcting the input schema stopped the Actor from auditing unintended packages, aligning audit outputs with caller intent and improving reliability.

1. Connect the Actor to Claude using the Apify MCP command with the ?tools parameter to expose only the desired tool.
2. Authenticate via OAuth; handle terminal requirement by running the CLI in a fresh console or batch file.
3. Send a minimal JSON input specifying only the desired repository.
4. Observe that the platform expands the input to include default packages due to the schema.
5. Identify the misconfiguration in the input schema (prefill vs default).
6. Update the schema to use "default" for packages, removing unintended prefilled entries.
7. Re-run the audit to confirm only the requested repository is processed.

The platform actually stored: ... "packages":["npm:request","npm:babel-eslint","npm:left-pad"] ...

The cause is one line in my own input schema, written months earlier.