Observability is not enough
Knowing what happened after an agent acts is useful, but teams also need ways to reduce what the agent could do in the first place.
Control before action
Capability pruning helps describe the pre-action control layer that narrows tools, context, and permissions.
Why the domain matters
CapabilityPruning.com gives this emerging topic a clear name for founders, open-source projects, cybersecurity vendors, and agent infrastructure teams building around AI agent permissions.