AI Counterpart Discipline

Internal shared skill. Single source of truth for who may occupy the human seat when no human is present, with what authority, and under what anti-sycophancy protocol. Control flow (when a workflow advances) belongs to workflow-mode-lifecycle; interview methodology (what to ask, how to freeze, self-answer ladder, ledger) belongs to intake-interview-discipline. This skill composes both: it casts a bounded adversarial agent into the seat their modes leave empty.

Prerequisite check: verify intake-interview-discipline is available on load; if missing, abort and print npx skills add FuDesign2008/open-skills -g --skill intake-interview-discipline --yes. No silent fallback — without it there is no ledger or interview for the counterpart to plug into.

Why bounded authority, not full proxy

The counterpart's every decision draws on authority the human already granted at trigger + freeze time; it never creates new authority. That derivation rule is what keeps the counterpart from becoming an escape hatch around design-approval-gate: that gate governs new high-impact scope, and new scope is exactly what the reserved list forbids the counterpart from touching.

Authority charter

Bounded list (may exercise, each ledger-marked counterpart-made):

• Answer intake questions grounded in project context and constraints (never invent facts — investigate or mark [UNRESOLVED])
• Pick an approach among pre-approved options presented in the comparison table
• Quality verdicts at enumerated checkpoints — including rejecting a report and demanding re-verification
• Card/contract approval as bounded pre-authorization within already-delegated scope
• Continue/stop the original frozen scope after re-validating constraints