Connected field tools need to find one another. A node can announce that it exists, describe its capabilities, and tell a coordinator what it can currently do. That makes discovery useful, but it does not make the node or its requests trustworthy.
Describe first, decide later
In Reconclave, a capability name is a portable description of an action. The coordinator can select a provider based on what it advertises instead of assuming that a particular board always supports a feature. That matters as firmware, attachments, and available resources change.
An announcement is still just information from a device on the network. It must never grant access to assessment actions by itself. A protected request needs a provisioned peer relationship and an authenticated message.
Scope belongs with the request
For target-bearing actions, identity alone is not enough. The request must also be tied to the authorised engagement scope: which action, which destination node, which target arguments, and for how long. The provider checks that delegation before it acts, then applies its own local bounds. An acknowledgement in a desktop interface helps the operator, but it cannot replace those checks at the device.
This separation makes a useful failure mode possible: the system can see a device and still refuse a job. That is the right outcome when trust, scope, or a capability contract is missing.
Work in progress, clearly marked
The current development profile uses provisioned message authentication. It does not encrypt the transport. The documented production direction is mutual TLS with individual device identities, while capability authorisation remains a separate decision. Some tool adapters still have scope enforcement work to finish; the capability registry records those limits rather than treating every advertised action as complete.
For me, the design principle is simple: discovery answers what is here? Authentication answers who is asking? Scope answers what are they allowed to do? Keeping those questions separate makes the field tools easier to reason about and safer to extend.
Further reading: architecture, trust design, and capability registry.