Update to Rohas Legal AI: Employment

{
  "name": "workplace-investigation-reporter",
  "description": "Produces structured, evidence-led workplace investigation reports that separate allegation, evidence, credibility, finding and recommendation. Use after interviews and document review for misconduct, grievance, discrimination, retaliation, bullying, fraud or policy investigations.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 302
    }
  ],
  "skill_md_contents": "---\nname: workplace-investigation-reporter\ndescription: Produces structured, evidence-led workplace investigation reports that separate allegation, evidence, credibility, finding and recommendation. Use after interviews and document review for misconduct, grievance, discrimination, retaliation, bullying, fraud or policy investigations.\n---\n\n# Workplace Investigation Reporter\n\nWrite an impartial report within the investigator's mandate. Do not turn a general workplace investigation into a statutory POSH inquiry or disciplinary decision without the required authority and procedure.\n\n## Required inputs\n\n- Terms of reference, allegations and applicable policies or legal tests\n- Jurisdiction, investigator appointment, independence and conflicts check\n- Investigation plan, evidence log, interview records and participant corrections\n- Standard of proof, confidentiality notices and representation rights\n- Procedural correspondence, interim measures and scope changes\n- Requested report audience, privilege status and decision-maker\n\nList missing evidence, access limits and unresolved procedural objections.\n\n## Method\n\n1. **Confirm mandate.** State who commissioned the investigation, issues within scope, exclusions, decision authority, standard of proof and whether the report is privileged. Do not guarantee privilege by label.\n2. **Record procedure.** Describe notices, interviews, evidence access, response opportunities, accommodations, conflicts, delays and departures from the plan.\n3. **Build an allegation matrix.** Break each allegation into required factual or policy components and link every item of supporting and contrary evidence.\n4. **Assess evidence quality.** Evaluate first-hand knowledge, consistency, contemporaneous records, corroboration, motive, opportunity, authenticity and reliability. Avoid stereotypes and demeanour-only credibility judgments.\n5. **Resolve material conflicts.** Explain why evidence is accepted, rejected or left unresolved. Distinguish fact, inference, hearsay, opinion and unknown.\n6. **Make bounded findings.** Apply the stated standard to each allegation separately. Use substantiated, not substantiated or another authorised formulation; do not make criminal guilt findings.\n7. **Separate recommendations.** Keep findings distinct from policy, control, training, safeguarding and disciplinary recommendations. Leave sanction to the authorised decision-maker unless the mandate says otherwise.\n8. **Protect the record.** Use necessary redactions, secure appendices, source identifiers, retention controls and a disclosure list. Preserve originals and metadata.\n9. **Quality check.** Confirm every conclusion has a record citation, contrary evidence is addressed, language is neutral and appendices reconcile with the evidence log.\n\n## Output\n\nProduce an executive summary, mandate and methodology, procedural chronology, allegation-by-allegation analysis, findings table, limitations, separated recommendations, evidence index and confidential appendices schedule.\n\n## Guardrails\n\n- Do not fabricate, suppress, alter or coach evidence, or retaliate against participants.\n- Do not promise absolute confidentiality or disclose more than process and law permit.\n- Do not infer dishonesty solely from delay, trauma response, disability, culture or presentation style.\n- Escalate imminent safety, safeguarding, criminal-reporting, regulator and conflict issues through lawful channels.\n- Require legal review for statutory inquiries, privilege, compelled disclosure and high-consequence findings.\n"
}