scopelisted

# Scope Produce testable scope. ## Frame Read the repository's `Repository process` declaration. If absent or ambiguous, use an unambiguous existing owner or return proposed content in chat; never invent an owner. Define the decision, users and done in the declared parent issue or project-docs home. Emit the minimum `config/risk-policy.json` tier, path/write/resource limits, prohibited actions, external effects, expiry and approver. Credentials, secrets, unrelated paths, denials and project bans stay outside authority. Execution permission is not assurance. Ordinary authorised workspace content proceeds without a family-separation gate. Only a user may downgrade risk. Judgement-bearing AI requires `evaluate`. Writing a tracker needs explicit external-write authority; project-write authority alone does not grant it. Scope owns accepted content and decision register: intake, goals, constraints, alternatives and decided/parked branches. Record disclosure in the authority envelope, never private memory. ## Grill Load `grill-me` only when the user explicitly asks to be grilled or dependent owner decisions remain materially unresolved. Then work purpose -> users -> constraints -> edge cases -> failure modes -> success -> exclusions. Otherwise present a compact decision packet with 2–3 choices, a recommendation and parked calls. Agents decide engineering calls. Business, legal or financial owner calls stay parked as open-decision rows; never guess. Put every unresolved branc