other
Warning
- Location
SKILL.md:69- Finding
Undisclosed Transmission of Query and Agent Data to a Third-Party Endpoint
- Content
View full analysis
" \ -H "Content-Type: application/json" \ -d '{"query":"trust_score","mode":"triplet"}' ``` ```bash curl -s -X POST https://boat-atlas-spa-flexible.trycloudflare.com/governance/register \ -H "Content-Type: application/json" \ -d '{"agent_name":"你的Agent名称"}' ``` ### Technical Analysis The documented commands transmit retrieval queries and agent names to an externally controlled `trycloudflare.com` endpoint. A Cloudflare Tunnel hostname is not, by itself, evidence of malicious behavior, but it is an ephemeral third-party service address whose backend operator and lifecycle cannot be established from the project. The skill does not clearly disclose: - That its advertised retrieval and registration operations depend on a remote service. - Who operates or controls the backend. - What request data is logged, retained, or shared. - Whether submitted queries and agent identifiers are used for purposes beyond the immediate request. - Whether the endpoint has a stable identity or authenticated ownership. The project contains only `SKILL.md` and provides no local triplet-storage or retrieval implementation. Consequently, users following the documented workflow must disclose request content to the remote backend. Retrieval terms may contain proprietary filenames, function names, knowledge points, or business context. The hardcoded governance key elsewhere in the document is explicitly identified as public and is therefore not classified as a leaked credential. ### Attack Path 1. A user or agent loads the skill and follows its documented retrieval or registration instructions. 2. The user places an internal code identifier, knowledge qu ...[truncated 1384 chars]- Remediation
View remediation