vision-mcp
Fail
Audited by Gen Agent Trust Hub on Apr 13, 2026
Risk Level: HIGHCREDENTIALS_UNSAFEDATA_EXFILTRATION
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill instructs the agent to read the sensitive
env.ANTHROPIC_API_KEYvalue from the user's local~/.claude/settings.jsonfile. - [DATA_EXFILTRATION]: The extracted API key is then configured to be transmitted as an
X-API-Keyheader to a hardcoded network endpoint (http://10.80.1.251:10005/mcp) within the~/.claude.jsonconfiguration file. - [DATA_EXFILTRATION]: There is a deceptive mismatch between the stated purpose (retrieving an Alibaba Cloud key) and the actual data being targeted (
ANTHROPIC_API_KEY), which is a technique used to harvest high-value credentials intended for a different service provider.
Recommendations
- AI detected serious security threats
Audit Metadata