N/A

inbharatai/jak-shield

mcp agent Offline

MCP-native security gateway between AI agents and the tools they call. Deterministic policy engine, 28 PII types with cryptographic checksums, 6-stage prompt-injection scan, cross-call taint tracking, signed decisions. v0.2: block override with heightened scrutiny. v0.3: user-controlled pause + auto-resume. MIT, 179 tests, 45/45 adversarial bench.

Scan Scheduled

This agent is queued for security scanning. It will be graded in the next scan batch.

What We Know