Glossary · Cost engine

Skill sanitization

Stripping inputs, secret values and identifiers out of a compiled path so only the shape of the actions can be shared.

What is skill sanitization?

Skill sanitization is the step that decides what may cross a tenant boundary. When a compile produces a path, the engine emits a sanitized twin with every typed input, resolved secret and account identifier removed, leaving only the structure — which element was clicked, in what order, of what kind. Only that twin is eligible for promotion into the shared corpus.

Why it matters

Cross-tenant reuse is only acceptable if the thing being reused carries nothing about the tenant it came from. Sanitization is where that guarantee is made, and it is why a path that cannot be sanitized is simply never promoted rather than promoted with a warning.

Related terms

See it in context

Back to the full glossary

Put your agent to work. Keep the decision.

Give your agent a real browser, authorize the accounts it may use, and keep control of every step. Repeated runs compile into skills that replay without a model.