agentward-ai/agentward

Adds three audit-facing features that close the gap between AgentWard's existing controls and what EU regulated entities (DORA, EU AI Act, MiFID II) actually have to put in front of an auditor.

EU AI Act framework (8 controls)

agentward comply --framework eu_ai_act

• eu-ai-act.art9-risk-management — zero-trust default action
• eu-ai-act.art12-record-keeping — audit trail availability
• eu-ai-act.art12-detection — sensitive content detection
• eu-ai-act.art13-transparency — registry cross-reference
• eu-ai-act.art14-human-oversight-write — gate write-capable AI components
• eu-ai-act.art14-human-oversight-network — gate network-capable AI components
• eu-ai-act.art15-cybersecurity — behavioral baseline
• eu-ai-act.art25-provider-chain — accountable owner & subcontractor disclosure

policy.skill_metadata schema (operator-supplied)

• SkillMetadata.owner — accountable party for a skill
• SubcontractorEntry — vendor / role / data_residency / contract_status

Surfaces in DORA Art. 28(3) and EU AI Act Art. 25 findings as RECOMMENDED gaps when missing. No auto-fix; the operator must declare these facts. DORA control count: 9 → 10.

agentward report (HTML Evidence Pack)

agentward report --frameworks dora,mifid2,eu_ai_act --audit-log audit.jsonl

• Single self-contained HTML file: no external CSS/JS, print-friendly via @media print rules.
• Sections: header, executive summary cards, policy summary table, per-framework findings, audit-chain integrity verification, scan inventory, footer.
• Verifies HMAC audit chain when a log is supplied; embeds the last 25 entries by default. Tampered chains surface as BROKEN with the first-break line number.
• Browsers handle the PDF round trip via Print → Save as PDF.

Test coverage

• 25 EU AI Act tests
• 8 subcontractor-chain tests across DORA + EU AI Act
• 16 Evidence Pack tests
• Full suite: 3,560 passing

Install

pip install --upgrade agentward