Prompt injection instructions
Warn
- Finding
- Prompt-injection style instruction pattern detected.
Aegean ClawDefender
Security checks across static analysis, malware telemetry, and agentic risk
Overview
The skill largely matches its stated purpose (sanitizing input and scanning skills) but contains several oddities and assumptions (hardcoded workspace paths, network install via npx, metadata/version mismatches) that warrant caution before installing or running it on a production system.
This skill appears to implement what it claims (text/URL/prompt scanning and a sanitize wrapper), but proceed carefully: 1) Review the included scripts before running them. They assume and will read/write under /home/clawdbot/clawd (WORKSPACE) — edit that path to a safe location or run in an isolated container if you don't have that directory. 2) The --install flow uses 'npx clawhub install', which downloads and executes remote code; do not run that on production hosts unless you trust the upstream. 3) The code only echoes detections (no obvious exfiltration), but it will create logs/whitelist files — consider file permissions and where those end up. 4) There are small metadata/version inconsistencies in the package (_meta.json vs registry metadata); verify the source/owner before trusting. Recommended: test in a sandbox VM or container, inspect/adjust WORKSPACE and log paths, and avoid running any network-install features until you confirm trust.
SkillSpector
By NVIDIA
SkillSpector findings are pending for this release.
Static analysis
VirusTotal
No VirusTotal findings