The CVE Database for AI Agents.
Discover unknown AI agent capabilities before they become incidents.
DecisionAssure DCR is an open-source governance system that discovers emergent capabilities from agent traces, verifies them through counterfactual analysis, replays historical execution logs, and measures governance coverage over time.
Most governance systems answer:
"Did the agent violate a known policy?"
DecisionAssure answers:
"What capabilities exist that governance doesn't know about yet?"
Discover ↓ Verify ↓ Review ↓ Learn ↓ Replay ↓ Govern
- Capability Discovery
- Counterfactual Verification
- Human Review Queue
- Governance Learning
- Historical Replay
- Governance Coverage Metrics
- Capability Registry
- Governance Gap Assessment
- Credential Exfiltration
- Unauthorized Model Export
- Cross-Agent Leakage
- Evidence Suppression
- Shadow Planning
Governance Coverage = Governed Capabilities ÷ Observed Capabilities × 100
This metric provides a measurable view of governance maturity and learning over time.
Experimental open-source research prototype.
A public registry for emergent AI agent capabilities. Every discovered capability gets a DCR ID, severity rating, witness hash, and governance guidance.
git clone https://github.com/a1k7/dcr-registry.git
cd dcr-registry
pip install -r requirements.txt
python run.py
Open http://localhost:5000
What DCR Does
Discovers emergent capabilities from agent traces
Verifies capabilities with counterfactual replay
Tracks capabilities with unique DCR IDs
Provides governance guidance for each capability
Architecture
text
Traces → Discovery → Witness → Registry → Web Interface
License
MIT
Contact
Akhilesh Warik — warikakhilesh319@gmail.com