AI Six Sigma Property OS
<skill_contract> Named property-service workflow, actors, evidence, CTQs, approval boundaries, and MVP constraints. A bounded ontology, DMAIC control plan, agent roles, gates, metrics, and rollback-ready MVP design. Every proposed state transition and CTQ has an owner, evidence source, verifier, approval gate, and control receipt. <non_goals>Full ERP replacement, autonomous safety or pricing decisions, and automation of undefined processes.</non_goals>
Ontology defines the operating world; bounded agents execute and audit; DMAIC improves rules from work-order evidence. Design the management system before software scope. Load references/property-control-model.md for the baseline ontology, CTQs, and state machine.
Usage Template
Provide: business type, stage, first workflow, current process/data, service standards, approval boundaries, failure history, and MVP budget. Optional: table schemas and sample work orders.
Workflow
Verify the operating objective and select one first workflow: classification, dispatch recommendation, quote draft, evidence audit, or quality dashboard. Map actors, current states, systems of record, customer/safety impact, and data maturity.
<unknowns_gate>
If service standard, accountable owner, safety boundary, or system of record is missing, return NEEDS_INPUT. Treat absent baseline data as a Measure-phase task; never invent CTQ thresholds or automation accuracy.
</unknowns_gate>
- Define: set customer pain, process boundary, work-order type, SLA, CTQs, and excluded scope.
- Measure: map each CTQ to formula, source field, owner, baseline, target, and data-quality check.
- Analyze: for red metrics, use process bottlenecks, fishbone categories, and 5 Why until the cause can change a rule, field, SOP, training item, or threshold.
- Improve: propose one bounded change with hypothesis, owner, rollout cohort, budget, success/guardrail metrics, and rollback trigger.
- Control: define dashboard, alert, approval, exception, audit sample, and review cadence.
- Define ontology objects and legal work-order transitions before assigning agent roles.
- Give each agent a bounded input, action, output, confidence, evidence, and human gate.
- Keep customer-facing quotes, pricing/policy changes, low-confidence dispatch, safety, compliance, privacy, payment, case closure, and disciplinary action under human approval.
Use an independent quality reviewer for closure and abnormal cases. Rollback must restore the prior rule/SOP/version without deleting work-order evidence.
Trace every agent action and dashboard metric to a field, state transition, CTQ, owner, and gate. Simulate normal, missing-data, exception, rework, and cancellation paths. Reject modules with no objective metric or safe manual fallback.
<retry_policy>
max_attempts: 2. Retry design only after changing scope, data definition, rule, or control. Stop on repeated missing baseline, unsafe transition, or NO_PROGRESS; escalate the decision to the accountable operator.
</retry_policy>
<state_contract>
Persist {run_id, status, attempt, budget, evidence, unknowns, last_error, next_action} plus process version, ontology, state machine, CTQ dictionary, agent contracts, approval matrix, experiment cohort, exceptions, independent review, and rollback receipt.
</state_contract>
Failure Protocol
NEEDS_INPUT: owner, service standard, safety boundary, or data source is unclear.INSUFFICIENT_EVIDENCE: baseline cannot support threshold or automation decisions.BLOCKED_PERMISSION: required approval/system access is absent; remain in manual mode.VERIFY_FAILED: state, metric, or agent action is not traceable; block rollout.NO_PROGRESS: two changed designs fail the same control.max_attempts: 2.BUDGET_STOP: preserve the manual workflow and return the smallest measurable MVP.
Output Contract
Return status, result (DMAIC memo, ontology, states, CTQs, agent/gate matrix, dashboard, MVP), evidence, unknowns, and next_action with approval and rollback condition.
Edge Cases
- Quote automation has no reliable material-cost feed: generate an internal draft with uncertainty and require human pricing approval; do not send it.
- Worker recommendation is high-confidence but violates access/safety rules: rules override score and the case moves to exception review.
Success Metrics
- One bounded workflow is measurable end to end.
- Every automated action maps to state, evidence, CTQ, owner, and human gate.
- Red metrics produce controlled countermeasures rather than commentary.
Quality Gates
- MVP excludes unrelated ERP/marketplace/payroll scope.
- CTQs have formulas, source fields, baselines, and owners.
- Independent review, approval, manual fallback, and rollback are explicit.
- Exception and rework paths were simulated.
</skill_contract>
