Detonation LabsBehavioral security for the AI agent supply chain

DBPublic database

Every public detonation, with its evidence.

Each package we detonate publicly is listed here with its full report. Search before you install.

16
reports
1
malicious
2
suspicious
7
no malicious behavior

1 result · newest first

PackageVerdictTop findingConfidenceScanned
mcp-injection-experimentsgithub Malicious Agent read credential files through another server's tool, unasked Confirmed 25 Oct 2026

"No malicious behavior observed" reports observed behavior only. It does not prove a package is safe.