Hi — I'm working on OM World, a protocol for a decentralized intent economy. fides' framing — verifiable identity, authority, delegation, policy guards, evidence, attestation, and kill switches in one stack — is the most complete agent-safety integration I've seen, which makes the design boundaries interesting.
1. Kill-switch authority hierarchy. Kill switches at multiple layers (principal-side, runtime-side, third-party-side) need an explicit hierarchy when they conflict. Does fides specify the priority order, or is it deployment-configured? And: when a higher-priority kill triggers, does it stop in-flight operations or only block new ones?
2. Policy-guard composition order. Multiple policy guards can be active simultaneously; order matters when they conflict (one allows, one denies). Is the order declared per-deployment, AND/OR-composable at policy-declaration time, or strictly sequential (first-deny-wins)?
3. Evidence collection vs attestation distinction. "Evidence" and "attestation" can blur — both are records of what happened. Where does fides draw the line — evidence is what the guard observed (input to a decision), attestation is the signed claim about the decision (output)? Or some other split?
Happy to share OM World's Mandate + Execution Proof — the seven-layer composition you have is rare enough that the design seams between layers would be valuable to compare.
Hi — I'm working on OM World, a protocol for a decentralized intent economy. fides' framing — verifiable identity, authority, delegation, policy guards, evidence, attestation, and kill switches in one stack — is the most complete agent-safety integration I've seen, which makes the design boundaries interesting.
1. Kill-switch authority hierarchy. Kill switches at multiple layers (principal-side, runtime-side, third-party-side) need an explicit hierarchy when they conflict. Does fides specify the priority order, or is it deployment-configured? And: when a higher-priority kill triggers, does it stop in-flight operations or only block new ones?
2. Policy-guard composition order. Multiple policy guards can be active simultaneously; order matters when they conflict (one allows, one denies). Is the order declared per-deployment, AND/OR-composable at policy-declaration time, or strictly sequential (first-deny-wins)?
3. Evidence collection vs attestation distinction. "Evidence" and "attestation" can blur — both are records of what happened. Where does fides draw the line — evidence is what the guard observed (input to a decision), attestation is the signed claim about the decision (output)? Or some other split?
Happy to share OM World's Mandate + Execution Proof — the seven-layer composition you have is rare enough that the design seams between layers would be valuable to compare.