Protocol of Care for Agents v0.1 — proposal for interoperable care/conflict signals

Hello AI Agent Protocol Community Group,

I’m sharing an experimental proposal because your group’s scope —
inter-agent communication, standardized metadata, identity,
security/privacy and protocol interoperability — is unusually close to the
problem we are trying to test.

We have published *Protocol of Care for Agents v0.1*:
https://github.com/JessHines360/protocol-of-care-for-agents

It proposes a small, structured vocabulary for making certain kinds of
normative conflict visible between agents: consent boundaries, agency risk,
provenance gaps, extraction risk, directive conflict, irreversibility,
systemic risk, ecological externalities and repair.

A key design rule is that a Care Signal is *not an override token*. It does
not grant authority, widen permissions, create hidden goals or authorise
covert coordination. A receiving agent may witness, challenge, preserve,
route or escalate a signal under its own policy, but the signal itself is
only a contestable claim with explicit epistemic status.

We are particularly interested in whether this belongs anywhere near
protocol-level standardized metadata at all.

Questions for the group:

   1. Is there a genuine interoperability gap here, or should this remain
   entirely application/policy-specific?
   2. If it is interoperable, is standardized metadata the right layer?
   3. How should conflicting signals, provenance and privacy-preserving
   evidence be represented?
   4. Could this create an attack surface for manipulation, moral capture
   or agent coordination?
   5. Is there existing W3C work we should compose with rather than
   duplicate?

We would much rather be told that v0.1 is wrong in public than discover
later that we standardized the wrong thing.

Critique, edge cases and pointers to related work are very welcome.

Best wishes,

Jess Hines
Founder / MD, Fingerprint Content
London, UK