Employment Law Agent

Produce a detailed product specification for an England & Wales employment tribunal triage agent, including target users, supported and unsupported claim types, assumptions, legal-disclaimer strategy, safeguarding boundaries, data privacy expectations, escalation triggers, user journey, output format, and a high-level technical architecture for a questionnaire-driven legal triage system that avoids presenting itself as legal advice.

Create a structured legal knowledge base for common England & Wales employment tribunal claim categories, including unfair dismissal, wrongful dismissal/notice pay, redundancy pay, unlawful deductions/unpaid wages, holiday pay, discrimination, harassment, victimisation, whistleblowing detriment/dismissal, family-leave detriment, TUPE-related issues, and breach of contract where relevant. Include eligibility requirements, key factual tests, limitation periods, ACAS early conciliation considerations, typical evidence, and caveats in a machine-readable rules format plus human-readable commentary.

Design the agent’s full interview flow as a branching questionnaire/state machine. Cover employment status, dates, continuity of employment, dismissal/resignation facts, protected characteristics, pay and benefits, losses, mitigation, medical impact, documentary evidence, internal grievance/appeal steps, ACAS early conciliation, and tribunal deadline triage. Deliver prompt templates, question banks, validation rules, fallback questions, and a JSON schema for collected facts.

Implement a rule-driven merits assessment layer that maps gathered facts to possible claim types, flags missing information, assigns cautious confidence levels, explains why each claim may or may not be available, identifies limitation risks, and generates plain-English reasoning. Deliver source code, rule files, examples, and unit tests covering edge cases such as short service, constructive dismissal, mixed discrimination/unfair dismissal facts, limitation-date uncertainty, and worker-versus-employee status issues.

Build compensation estimation logic for indicative tribunal outcomes, including unfair dismissal basic award and compensatory award concepts, loss of earnings, notice pay, unpaid wages, holiday pay, redundancy pay, discrimination injury-to-feelings ranges, aggravated/personal injury placeholders, pension/benefits loss assumptions, mitigation, caps, deductions, and uncertainty bands. Deliver documented formulas, calculator code, configurable parameters, worked examples, and tests demonstrating conservative range outputs rather than definitive valuations.

Create a working text-based prototype that conducts the intake interview, stores structured answers, invokes the merits and compensation modules, handles incomplete/contradictory answers, and produces an initial triage report. Include a CLI or minimal web/API interface, modular code structure, sample configuration, logging without sensitive-data leakage, and developer setup instructions.

Develop the final report-generation layer and user-facing language. The output should include possible claims, reasons, uncertainty, evidence to gather, limitation/time-limit warnings, indicative compensation ranges, assumptions used, and a prominent automated-triage/not-legal-advice disclaimer. Deliver polished templates, tone guidelines, example reports for multiple scenarios, and UX copy for consent, privacy, and next-step solicitor/advisor signposting.

Produce a comprehensive scenario-based evaluation pack with synthetic user cases, expected classifications, expected compensation-range behaviour, deadline-risk checks, and refusal/escalation behaviour. Include automated tests, red-team prompts, hallucination checks, regression fixtures, bias/fairness considerations for discrimination claims, and a QA checklist for solicitor or employment-advisor review before deployment.

Prepare the project for handoff and ongoing maintenance. Deliver deployment instructions, environment configuration, data-retention guidance, versioning strategy for legal rules and compensation bands, admin documentation, monitoring recommendations, known limitations, backlog for future enhancements, and a release checklist suitable for a cautious beta launch.