Human Supervision & Escalation

Questions or comments?

Post them as a comment on the tracking issue -- requires a GitHub account with access to this repo.

Scope

Every Agent Worker must report to a named human employee — structural, not optional; no agent operates without a human accountability chain. Minor policy deviations are surfaced to the responsible human employee for correction, keeping accountability close to the work. Major violations are escalated through standard HR escalation paths, with a full audit trail.

Depends entirely on the policy-violation taxonomy (minor vs. major) being defined — per ../schema-changes.md, this is a P0 TODO owned by Product + Legal and explicitly Not Started in the source paper. This spec describes target behavior for both severities; it cannot be built as described until that taxonomy exists.

Value

Keeps accountability close to the work for routine issues (fast correction by the responsible human) while ensuring major issues reach the same escalation paths HR already trusts for human-employee issues — one consistent standard, not a parallel, agent-only process.

Acceptance criteria

• Every Agent Worker has exactly one named human supervisor at all times — this is enforced structurally, not left as an optional field.
• A minor policy deviation (per the taxonomy, once defined) routes to the responsible human employee with enough context to correct it.
• A major violation (per the taxonomy, once defined) escalates through the same standard HR escalation paths used for human-employee issues, with a full audit trail of the escalation.
• The distinction between "minor" and "major" is applied consistently and transparently — a human reviewing the audit trail can see why a given violation was classified the way it was, not just the classification itself.