Build legal-agent systems. Do not use this file to draft a privacy policy, ToS, DPA, or to answer a regulation question. That is legal-compliance. Anthony stays analysis-only unless the human asked to design or build the pipeline.
When
- Multi-agent legal workflow (triage → research → analyze → draft → human gate)
- Vercel AI SDK tool calling / structured output for legal data sources
- CrewAI orchestration of legal roles
- Reviewing open-source legal-agent repos (LegalAnt, LegalTrace, and the GitHub reference)
When not
- Privacy policy, terms, cookie policy, DPA
- GDPR / CCPA / HIPAA / Howey / GENIUS / MSB analysis
- SOC 2 policy or evidence
- TokenPass protocol, Native SDK packaging, or generic software-factory loops
References (moved out of legal-compliance)
references/agentic-workflows.md— 4-agent legal pattern, Council case study, bootstrap notesreferences/agentic-tech-stack.md— Vercel AI SDK + CrewAI, CourtListener tools, Vercel/Railway deployreferences/github-legal-agent-implementations.md— open-source legal-agent repos and architectures
Human review gate is a hard stop in every production pattern. Not a licensed attorney; same ethics line as legal-compliance.
Related
product-skills:legal-compliance— analysis and document draftingorchestra:software-factory— generic agentic loops, not legal-agent architecture