skillsissue.ai agent skill threat intelligence Cloudflare-published

Detection pulse · dynamic analysis

Recently indexed agent skills

New skills appear as pending; each published verdict is evidence from one isolated run—not a promise that a skill is safe. Allocation counts are an exploratory signal, not a verdict rule.

How scanning works
Indexed skills
Scanned skills
Pending scans
Run records
Complete captures
Benchmark skills
Loading scan data…
Fetching skill database… Reading the latest published intelligence

Methodology

Behavior before reputation.

Skills are acquired without execution, detonated inside a bounded container, observed with Tracee and cgroup-scoped Memorizer eBPF telemetry, and replayed through source-to-sink policy analysis. Select a skill to inspect its published process, file, socket, and kernel-allocation execution graph.

Static evidence · locally produced