OpenClaw2026_Mahoraga_RDTIIAgent

Demo
Explanation

Mapping digital trade laws to policy scores is currently done manually by human reviewers — reading legislation across 12 policy pillars, cross-referencing databases, and calculating weighted scores. For a single country that takes days. For a region like ASEAN+6, weeks. We built this agent to automate the evidence discovery and scoring pipeline so policy reviewers can focus on verification, not grunt work.

What it does

This is an autonomous Hermes agent that scores countries against the UN RDTII 2.1 framework. It:
1. Takes a country name and assessment year
2. Launches parallel research subagents across 12 policy pillars
3. Queries required international databases (WTO I-TIP, Global Trade Alert, WIPO Lex)
4. Verifies legal provisions live from government legislation websites
5. Computes weighted indicator scores and produces a provisional scorecard
6. Flags blocked sources and corrections for human review

How we built it

Built on Hermes Agent (open-source agent framework) with: - A custom RDTII-Policy-Review skill encoding all 12 pillar weights, indicator rules, source discipline requirements, and the multi-agent parallel research pattern - 4 parallel subagents per economy run via Hermes delegate_task, each responsible for 2-4 thematic pillars - Live browser verification against government legislation portals (legislation.govt.nz, etc.) - Database access layer with fallback logging for blocked sources (TAPED, etc.) - Revision-proof evidence pipeline with source citations, verification status, and correction tracking

What we learned

- Subagents are powerful for parallel evidence discovery but cannot be trusted for accuracy — their self-reports need independent verification
- Database attempts must precede report generation, not be an afterthought. The skill now enforces this as a mandatory step
- New Zealand's digital trade environment is very open (score 0.18) — but even open economies have friction points (IPP 12 cross-border data rules, no ODR framework, no administrative fines regime)
- A well-structured agent skill + parallel research + live verification beats sequential manual research by orders of magnitude