Agent Spec: Frontline Assistant (task reminders, SOP retrieval & lightweight coaching)

Purpose

This lightweight frontline assistant helps staff get the right instruction at the right time, reduces repetitive questions, nudges routine work (for example sanitation rounds and closing checks), and captures simple structured observations without taking autonomous corrective action. The agent is explicitly designed to support staff judgment, improve compliance with standard work, and create a searchable operational log for later analysis.

Scope

Primary capabilities delivered by the agent:

• Fast SOP retrieval by role, tag, or keyword (example: "server closing checklist").
• Push scheduled or ad-hoc reminders (example: sanitation rounds, temperature checks, prep lists).
• Micro-coaching prompts for common quick observations (example: plating consistency reminder during dinner rush).
• Capture simple incident reports and append them to the shift log (structured form).
• Role-aware responses and restricted actions (manager confirmation required for corrective steps).

Behavior and UX Patterns

• Primary channel: in-app chat. Secondary channels may include SMS or Slack for managers (future integration).
• Role-aware greetings and suggestions: agent greets staff by role and lists relevant quick actions ("See today's closing checklist, start sanitation timer, report an incident").
• When a user requests an SOP, show the short quick-steps up front with a link to full SOP and version metadata. Ask whether they want to mark steps complete or open the full procedure.
• For reminders, allow opt-in snooze and completion confirmation. Do not auto-reschedule without manager approval.
• Keep interactions short; provide explicit options (buttons or quick replies) to reduce typing during busy shifts.

Conversation Examples

• User: "Show me the server closing checklist."
  Agent: "Server Closing Checklist (v3.1) — ~6 items. Quick view: [list brief items]. Mark items done, view full SOP, or assign to another role."
• User: "I noticed a temperature issue in walk-in."
  Agent: "Please confirm: record incident for shift log? (Yes / No). If yes, I'll ask a couple quick fields: observed temp, location, immediate action taken. A manager will be notified to confirm corrective actions."
• Agent-initiated reminder: "Sanitation round due in 10 minutes. Start now?" with buttons: Start / Snooze 10m / Done.

Capabilities (detailed)

1. SOP Retrieval
   • Search by role, tag, keyword, or common phrase.
   • Return a short quick-steps summary (2–8 bullet steps) with link to full SOP and last-reviewed date.
   • Support versioning and show "effective from" date.
2. Scheduled & Ad-hoc Reminders
   • Configurable reminder templates: interval (e.g., sanitation every 2 hours), pre-close reminders, temp log prompts.
   • Allow staff to mark complete, snooze, or request manager attention.
3. Micro-Coaching Prompts
   • Short, contextual tips triggered by time, role, or user request (e.g., plating consistency tips during rush).
   • Non-punitive tone. Include optional quick checklists and offer to schedule a short follow-up coaching huddle.
4. Incident Logging
   • Capture structured incident entries for the shift log (see Data Fields below).
   • Attach optional photo or POS/order reference when available.

Data Requirements & Access

Minimum data sources the agent must access:

• SOP library with metadata (role tags, version, tags, full text) — read access.
• Shift roster and current active shift context (user role, location, manager on duty) — read access.
• Incident-form endpoint or shift log API for structured submissions — write access.
• Reminder / schedule store for per-shift reminder state (complete/snoozed) — read/write.

Suggested Incident Data Fields

• timestamp (auto)
• reporter_name / reporter_role
• location / station
• short_description (text)
• severity (low / medium / high)
• immediate_action_taken (text)
• manager_notified (yes/no), manager_name
• optional_attachment_reference (photo id)

Safety, Permissions & Rules

• Agent must never perform corrective actions autonomously (no schedule changes, no inventory adjustments, no disciplinary messages). All corrective or schedule-changing recommendations require explicit manager confirmation and an auditable manager action recorded.
• Sensitive actions (e.g., closing shift early, changing posted menus, discarding product) must be gated and require a manager override recorded in the log.
• Access control: role-based access to SOPs and to incident submission features. Managers may have additional capabilities (view full shift log, acknowledge incidents, close corrective items).
• Privacy: do not collect personal medical data without separate consent and policy. Limit PII collection to necessary fields for operational follow-up.

Integrations & APIs

Recommended integrations for full value:

• SOP library API (search, fetch summary, fetch full SOP).
• Roster / scheduling system (active shift context).
• Shift log / incident endpoint (POST structured incidents).
• Optional: POS for order linking, temperature-monitoring systems for automated alerts, messaging channels for manager escalation.

Pilot Plan

1. Scope: Single high-volume location, one month pilot (4 weeks).
2. Initial goals: improve closing checklist completion rate, increase sanitation-round compliance, and reduce time-to-report incidents.
3. Baseline measurements (week prior): closing checklist completion %, sanitation rounds completed per shift, average incident report time, staff satisfaction baseline survey.
4. Pilot rollout steps:
   1. Integrate SOP library and roster data.
   2. Configure 3–5 reminder templates (closing checklist, sanitation rounds, temperature checks).
   3. Train staff with a 30-minute demo and quick reference card.
   4. Run pilot, monitor for 4 weeks, collect usage logs and staff feedback weekly.
5. Success criteria (examples):
   • Closing checklist completion rate improves by 20 percentage points or reaches ≥90%.
   • Sanitation rounds completion rate increases by 30%.
   • Average incident report latency decreases by 50%.
   • Staff survey shows neutral-or-better sentiment and decreased time spent looking for SOPs.

Monitoring & Continuous Improvement

Collect and review:

• Agent usage (commands used, search queries, reminders acted on, incident submissions).
• False positives/negatives in SOP retrieval (misleading suggestions) and friction points reported by staff.
• Manager override frequency and reasons.

Implementation Checklist

1. Confirm SOP library API and roster access.
2. Create initial reminder templates and assign owner for each template.
3. Build incident submission schema and test POST endpoint.
4. Define role-based access rules and manager override workflow.
5. Prepare staff briefing materials and brief managers on escalation policy.

Future Enhancements (not required for pilot)

• In-app checklist item completion tracking with saved state across shifts.
• Low-friction voice interface for hands-busy environments (requires separate safety review).
• Analytics dashboard combining POS, incident, and compliance data to identify hotspots.
• AI-assisted suggestion engine to highlight SOPs most relevant to frequent issues (requires careful guardrails to avoid overreach).

Design note: keep the agent helpful and lightweight. Avoid turning it into a policing tool — its value is in helping staff do their work faster and more consistently, and in creating a reliable record that managers can use to improve systems, not punish people.