Licensing Review JSON

Licensing Review JSON

Use this skill for licensing tasks that provide a base URL, REST endpoints or SQL access, and an input/payloads/answer_template.json schema. The job is to derive structured review results from the data service and return only the JSON object required by the template.

Core Workflow

1. Read the user prompt and the answer template before querying data. Extract target IDs, review date or boundary date, queue size, required top-level keys, allowed enum codes, and ordering rules.
2. Resolve <TASK_ENV_BASE_URL> only from the environment access instructions provided with the task. If SQL is allowed, include the required token/header from those instructions.
3. Fetch /api/policies and the task's listed business endpoints. Prefer SQL for complete target-specific record pulls when it is available, because public GET lists may be capped or ordered so target records are missing.
4. Parse embedded JSON fields such as details_json and controls_json.
5. Build an intermediate worksheet keyed by the stable target ID. Keep notes internally, but the final response must be only JSON matching the template. Never include markdown, citations, comments, or extra keys.
6. Normalize output to the template: use only allowed enum values, empty arrays when no values apply, de-duplicate arrays, and apply every ordering rule exactly.

SQL Patterns

Use table names that mirror endpoint names, for example policies, contractor_applications, contractor_bonds, contractor_insurance, contractor_license_history, contractor_violations, contractor_correspondence, contractor_inspections, liquor_applications, liquor_settlements, liquor_privileges, liquor_incidents, liquor_site_evidence, alcohol_licensees, alcohol_violations, and renewal_rules.

Issue focused queries for the target IDs or locations, not broad table dumps. If a SQL endpoint returns truncated: true, add narrower where clauses.

Contractor Batch Decisions

Use this workflow when the schema has application_decisions and contractor deficiency/action fields.

Data joins:

• Target applications come from the prompt and template.
• Match current contractor policies by family = contractor, trade, and requested class. Use the latest effective policy on or before the review date. Use legacy contractor policy only for policy_impacted.
• Join bonds, insurance, correspondence, and inspections by application_id.
• Join license history by prior_license_id.
• Join violations by related_application_id; also include rows for the prior license when that relationship is material.

Deficiency rules:

• Bond: active only means status = active and no cancellation before the review date. If no active bond exists, use the template's no-active-bond code such as no_active_bond or bond_cancelled. If an active bond exists but its amount is below policy minimum, use the template's bond-shortfall code.
• Insurance: require an active policy, amount at least the policy minimum, and expiration on or after the review date. Pending insurance is not current; use insurance_pending when available, otherwise insurance_not_current. Expired active insurance uses insurance_expired when available. Low active coverage uses an insurance-shortfall code.
• Endorsement: if the current policy requires an endorsement, missing and pending statuses are deficiencies. Use exact codes such as endorsement_missing or endorsement_pending when available; otherwise use the template's not-verified code.
• Experience: years of experience below policy minimum is an experience-shortfall deficiency.
• License history: a suspended prior license is an active_suspension deficiency.
• Violations: open serious violations or complaints block approval and use the template's serious/open complaint code. Open minor violations are only output when the template has an open-minor code. Dismissed or resolved rows do not create application deficiencies.
• Inspections: failed or conditional DOC_GAP findings create an inspection document-gap deficiency when the template supports it. Failed SAFETY_RECHECK findings create a safety-recheck deficiency when supported. Passed findings are not deficiencies.

Required actions should follow the deficiency using the action vocabulary available in the template. Typical pairings are:

• bond shortfall: increase bond amount.
• no active or cancelled bond: file or obtain a current bond.
• insurance expired or not current: provide or renew current insurance.
• insurance pending: verify insurance binding.
• insurance shortfall: increase insurance amount.
• endorsement missing: obtain required endorsement.
• endorsement pending or not verified: verify endorsement.
• experience shortfall: submit or document experience evidence.
• active suspension: board review and clear or review the suspension.
• open serious violation/complaint: resolve the violation or complaint and send to board review when that action exists.
• open minor violation: resolve minor violation review.
• inspection document gap or safety recheck: clear the document gap or complete the safety recheck.

Determination and risk:

• DENY when an active suspension or open serious violation/complaint exists.
• HOLD when there is any deficiency but no denial blocker.
• APPROVE only when no deficiencies remain.
• Risk is high for denial blockers, medium for holds, and low for approvals unless the prompt/template defines a stricter tier rule.

policy_impacted is true only when a current policy threshold or endorsement rule creates a material deficiency that would not have applied under the prior baseline. Do not mark ordinary stale documents, expired coverage, suspensions, or open violations as policy impacts by themselves.

The summary counts must be computed from application decisions. High-risk and policy-impacted ID lists are sorted. For stale or unverified correspondence, include correspondence that is not agency-verified or is stale/conflicting on its face, unless another field clearly says it was verified by licensing staff.

Restricted Liquor Staff Packages

Use this workflow when the schema asks for recommended_posture, same_premises_basis_applies, risk/control code arrays, a first-90-day plan, and escalation triggers.

Data joins:

• Select the target liquor application and target location from the prompt.
• Join settlements, incidents, and site evidence by location_id.
• Load privilege rows for the application's license_class.
• Current location-specific controls are the union of controls in active, unexpired settlement controls_json.
• Standard obligations are privilege rows with standard_required = 1; keep these separate from location-specific controls.

Field rules:

• same_premises_basis_applies is true when same-premises settlement history exists for the location and the prompt asks whether that basis remains applicable; otherwise rely on the active current basis.
• covered_risk_codes should reflect risks actually addressed by current controls or current active settlement bases. Translate controls where the template uses risk vocabulary, such as patio controls to patio-boundary risk. Do not include dismissed incidents as covered risks unless a current control is explicitly tied to that risk.
• Include open or referred incident risks when current controls address them; otherwise they usually become verification gaps.
• verification_gap_codes come from missing, conflicting, stale, or identity-mismatched evidence; open/referred incidents; unresolved tax holds; and prompt-specific verification needs such as camera, food-service, floor-plan, signage, police-memo, neighbor-notice, site-photo, or late-night monitoring.
• recommended_posture is request_follow_up when verification gaps or unresolved incidents remain, deny for unresolved major blockers that cannot be conditioned, and issue_restricted when current controls cover the risks and no material verification gaps remain.

Common evidence and gap mappings:

• CONTROL_SIGNAGE missing/conflicting: use the template's control-signage missing/conflicting code.
• FLOOR_PLAN conflicting/stale: use the floor-plan conflicting or stale code.
• POLICE_MEMO conflicting or identity-mismatched: use the police-memo code when available.
• Missing camera/CCTV evidence: use the template's camera/CCTV evidence gap and include a camera walkthrough/export check when available.
• Missing food-service evidence for a class with a food-service standard: use the food-service evidence gap and include a food-service check when available.
• Open tax hold: use the tax-hold gap and tax-clearance or tax-hold escalation code when available.
• Open or referred minor-sale incident: use incident follow-up and referred-minor-sale escalation codes when available.

For the first_90_day_plan, pick checks that directly test unresolved gaps and active controls. Evidence and record verification usually belongs in first_30_days; late-night or after-hours field visits usually belong in days_31_60; noise/patio boundary monitoring often belongs in days_61_90. Follow any ordering rule in the template; otherwise use operational sequence.

Escalation triggers should mirror the risk and gap set: after-hours service, unverified control signage, major incidents, unresolved minor-sale referrals, camera/security failure, missing footage, missing food service, noise/patio breach, and uncleared tax holds, using only codes allowed by the template.

Alcohol Renewal Manual-Review Queue

Use this workflow when the schema has a ranked queue and summary for alcohol license renewals.

Data joins and filtering:

• Target active licensees come from the prompt and template. Ignore inactive licensees except when an active target has a successor_to license.
• Use the boundary date from the prompt; if absent, select the matching renewal_rules row and use release_boundary or details_json.use_violations_on_or_before.
• Include exact violations for the current license_no with violation_date <= boundary.
• If the target has successor_to, include predecessor violations only when the predecessor record is an unambiguous address/facility continuation. Mark match_confidence as close_address for a clean predecessor match and uncertain when names or addresses conflict.
• Exclude violations after the boundary, but list their IDs in summary.post_boundary_violation_ids_excluded sorted by ID.

Queue fields:

• matched_violation_ids: sort by violation date ascending, then ID ascending.
• violation_count: count only matched pre-boundary violations.
• most_recent_violation_date: latest date among matched violations used for ranking.
• match_confidence: exact when only current-license exact rows were used, close_address for clean successor/address matches, and uncertain for ambiguous matches.
• next_step_label: use board_review for successor/close-address serious matches, unresolved serious control risks, or repeated high-risk themes such as sale-to-minor, tax-hold, or after-hours issues. Use manual_fine_check when unpaid fines or positive balances require a hold and board review is not required. Use manual_ALERT_check when alert flags require manual review but no stronger label applies. Use additional_record_check for uncertain matches or incomplete identity evidence.
• risk_tier: high for board review, unresolved serious violations, significant unpaid-fine review, or repeated violations; medium for alert-only or older lower-severity queues; low only for minimal residual manual checks.

Ranking:

1. Build candidates with at least one matched pre-boundary violation or other rule-triggering manual-review reason.
2. Assign action priority: board_review, then manual_fine_check, then manual_ALERT_check, then additional_record_check.
3. Within the same action priority, sort by most recent matched violation date descending, then violation count descending, unresolved serious count descending, positive fine balance descending, and license number ascending.
4. Emit ranks 1..N with no gaps and truncate to the requested queue size.

The summary queue size must equal the emitted queue length. close_or_uncertain_match_license_numbers contains every queued license whose confidence is not exact. board_review_license_numbers contains every queued license with next_step_label = board_review, sorted by license number.