Policy Verdicts

Posture: docs_first, advisory_first. Describes vocabulary and required behavior; implements no runtime, permission engine, or tool. Provider-agnostic.

Goal

Name the small set of outcomes a permission decision can produce, so skills, contracts, and audit records share one vocabulary. A verdict is what the harness decides about a proposed action — it is decided outside the model and recorded.

Verdicts (declaration vocabulary)

policy_verdicts:
  allow:           { meaning: "Action may execute." }
  deny:            { meaning: "Action must not execute." }
  require_approval:{ meaning: "Action pauses until explicit human approval." }
  transform:       { meaning: "Action may execute only after being reduced, masked, scoped, or converted." }
  log_only:        { meaning: "Action may execute but must be logged due to a traceability requirement." }

These five are the vocabulary a skill or reviewer uses to talk about outcomes (e.g. a restricted_tools[].restriction in a per-skill harness_requirements).

Reconciliation with the harness decision values

The runtime harness emits a richer set of decision values, defined authoritatively in agent-harness-governance-extension.md and tool-permission-matrix.md: allow, deny, ask_user, approval_required, require_stronger_auth, run_in_sandbox, run_as_draft_only.

The five verdicts above are a projection of those values for declaration and review. Map them:

Verdict (declaration)	Harness decision value(s) (enforcement)
`allow`	`allow`
`deny`	`deny`
`require_approval`	`approval_required` (or `require_stronger_auth` for high-risk)
`transform`	`run_as_draft_only`, `run_in_sandbox` (scoped/converted execution)
`log_only`	`allow` with `audit_policy: full_structured_event`

ask_user (clarification before deciding) has no declaration-level verdict; it is a harness behavior that precedes a verdict. When declaration and harness disagree, the harness decision values govern; the verdict vocabulary is for human-readable declaration and audit.

Policy Verdicts

Goal

Verdicts (declaration vocabulary)

Reconciliation with the harness decision values

See also