Schema changes for agent-accountability-framework¶
AI-DRAFT, higher uncertainty than this repo's other two schema deltas. Neither
../../schema/nova-advisor-sn-app.mdnor../../schema/nova-codex-curator.md— the two schemas read for this repo — contain anything resembling an Agent Worker, Job Profile, or policy-violation record. The source concept paper's own P0/P1 TODO list confirms this: "Build the Job Profile schema for Agent Workers" and "Define the policy violation taxonomy" are both listed as Not Started, owned by Product (and Product+Legal), respectively. This delta is therefore a placeholder naming what will need a schema, not a proposal ready for the schema-first gate.
What already exists¶
Nothing found in the two schemas reverse-engineered for this repo. The
concept paper itself doesn't name which Nova app/repo would own this data —
"Novaworks" broadly, and "ServiceNow" for the machine-identity side, are the
only systems named. Given the existing precedent of sys_user-style worker
records and x_novaw_platform_* cross-app tables referenced in
specs-superagent's own schema docs, an Agent Worker identity most likely
belongs in a platform-scope app (x_novaw_platform) rather than either of
the two Advisor-domain apps this repo's schema docs cover — but that's an
inference, not a confirmed placement.
Likely-needed delta (naming the shape, not committing to it)¶
- Agent Worker identity record — likely referencing (or extending) a
ServiceNow machine-identity concept, plus a link to the Functional Agent
or agent runtime it represents (see
specs-superagent/product/glossary.md's Functional Agent (FA) entry — this Feature Set's "Agent Worker" is explicitly a governance wrapper around an FA/agent runtime, not a replacement for it). - Job Profile record — role, scope, permitted actions, guardrails;
authored collaboratively by business leaders and HRBPs. Needs a schema
for both the profile content and its authoring workflow state
(draft/review/approved, mirroring the Policy Editor's draft→publish
pattern in
../nova-policy-advisor/native-policy-editor-collaborative-authoring/spec.md— worth checking whether these two authoring workflows could share infrastructure rather than each inventing one). - Human supervisor link — Agent Worker → named human employee, structurally required (not optional) per the source paper.
- Audit trace / policy compliance audit record — ingested from a LangFuse (or equivalent) connector; audited against the Job Profile's policy requirements. Depends on the "Spec the LangFuse connector and the audit ingestion model" P0 TODO, not started as of the source paper.
- Violation record — minor vs. major, per a taxonomy explicitly not yet defined (P0 TODO, Product + Legal). No schema should be committed here before that taxonomy exists — this delta intentionally stops short of proposing violation-record fields until that dependency clears.
- Performance rollup link — audit outcomes + violation history +
business value delivered, surfaced into the responsible human's
performance review cycle. Depends on Novaworks's existing performance-
monitoring data model (out of scope for this repo — see
specs-superagentfor that domain's schema, if relevant).
Given how much of this depends on not-yet-started specs (the taxonomy, the connector, the Job Profile schema itself), do not treat this file as ready for the schema-first gate — it exists to name the shape of work ahead, not to propose a mergeable delta. Revisit once the P0 TODOs in the source paper are closed.