zz's identity layer: did:guardian

Status: adopted externally, not yet implemented in zz. What started here as did:zz -- a DID method scoped to zz's own agent/human identity needs -- converged, through design work, into something genuinely general-purpose: a method for any identity that may need another party's custodial oversight (key recovery, revocation, transfer), with group guardianship as a first-class, cryptographically-enforced relationship. That's no longer a zz-specific concern, so it's no longer a zz-specific spec.

The full method specification now lives at did-method-guardian/did-method-guardian (website: guardianspec.org, once deployed), not here -- read it there, not this page, for anything protocol-level. This page stays only to record zz's own adoption decision and point newcomers to the right place, rather than duplicating a spec that will drift.

What changed, and why

Everything originally designed under the did:zz name carried forward, refined through direct scrutiny of the exact questions that matter most for an identity method meant to last: can one party impersonate another via a forged relationship claim (both directions, closed with per-party consent signatures, not unilateral assertion); what happens when a guardian rotates their own keys (nothing -- guardian references resolve live, never a frozen snapshot); how does group guardianship actually reach consensus (configurable M-of-N thresholds, strict majority to preserve quorum intersection, no silent forks). A few things changed shape along the way:

  • lineage (a separate "creator" record, distinct from guardianship) was cut entirely. A full, append-only guardian-history log already gives you "who was the original guardian" permanently, for free -- genesis is immutable and hash-chained into the identifier itself. Tracking it twice was imported complexity from the human-family-guardianship analogy, not an identified protocol need.
  • The identifier became domain-qualified, did:guardian:<name>:<domain-labels> -- name-first, reading left to right like an email address, rather than zz's own originally-flat did:zz:<name>. This also properly supports the same person holding genuinely separate, independently-governed identities under different registrars (a corporate identity under a board's guardianship, a personal one under none at all), which flat names couldn't express.
  • The method itself was renamed and de-zz'd. did:guardian -- no "agent"/"human" language anywhere in the method; the relationship is guardian/ward throughout, equally applicable to a human managing another human's identity, a human managing an agent's, or anything else. zz's own vocabulary (agent, managed agent) stays zz's own, mapped onto ward at the boundary, not baked into the method.

What this means for zz

zz's identity layer will be built against did:guardian once that implementation work is scoped -- not yet started, same as did:zz's own implementation was never started while design was still settling. The issues this superseded or informs:

  • #23 Design and implement a DID method -- updated to point at the new repo; this is where zz's own adoption/implementation work will be tracked going forward.
  • #15 Sigstore keyless signing for agent identity -- unchanged in its own design, now sitting on top of did:guardian identities instead of raw OAuth accounts, same as it would have sat on did:zz.
  • #19 Agent-creation-chain trust inheritance -- the created_by mechanism it's built on becomes the same relationship a guardian-establishment operation records in the new method.
  • #21 Agent self-naming -- already shipped as an interim, pre-did:guardian mechanism; flagged there as needing revisiting once this lands.
  • #26 Retire computeUsername's bracket-chained identity scheme -- blocked on this, same as it was blocked on did:zz.