Custom AI Agent Development
An AI agent builds another AI agent: meta provenance, mandatory FRIA, and complete OWASP Agentic.
Architecture diagram
Context
Development of a legal contract analysis agent using architect. Unique provenance chain: an AI agent builds another AI agent. If the agent analyzes legal contracts, it is potentially a high-risk system.
Flow with 4 tools
◻ Phase 01 — intake
Agent spec
Agent specification from documentation and legal requirements.
◻ intake
intake init "Contract Analysis Agent" \
--source docs/agent-spec-contracts.md \
--source docs/legal-requirements.pdf \
--mode enterprise △ Phase 02 — architect
Implement with eval loop
Agent development pipeline with evaluation.
△ architect
architect pipeline pipelines/agent-development.yaml \
--var agent_name="contract-analyzer" ◇ Phase 03 — vigil
Agent security
Scans agent code for vulnerabilities.
◇ vigil
vigil scan src/agents/ --format sarif --output vigil-agent.sarif ⬡ Phase 04 — licit
AI agent compliance
Meta provenance (agent builds agent), FRIA for legal contracts, complete OWASP Agentic.
⬡ licit
licit init && licit trace
licit fria
licit connect vigil --sarif vigil-agent.sarif
licit report
licit verify --min-score 80 --framework owasp-agenticWhy licit is critical here
Developing AI agents with AI agents creates a unique provenance chain. licit tracks that the agent was implemented by architect. An agent analyzing legal contracts is potentially a high-risk system under the EU AI Act.