confidential-data-handling

What problem does it solve?

This Skill prevents sensitive Wolfram Language data from being exposed in Claude prompts by guiding you to correctly mark inputs and outputs as confidential or to intentionally declassify them when needed.

Core Features & Use Cases

• Confidential wrapping at the right time: Wrap values during assignment using Confidential[...] so secret content is excluded from downstream prompts and propagation happens correctly.
• Automatic handling of secret dependency structure: When schema sending is enabled, confidential-dependent outputs include safe structural metadata (type/keys/sizes) instead of raw sensitive values.
• Declassification via NonConfidential: Produce non-confidential results even when they depend on confidential variables, skipping confidential propagation for the marked result.
• Trigger-based behavior: Use Confidential[...] when the user instruction includes “secret/confidential/hidden” in a variable-assignment context.

Quick Start

Mark a value as confidential by assigning it as var = Confidential[expr], or declassify a computed summary as summary = NonConfidential[... ].