jfrog-ai-catalog-skills

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). The runtime workflow reads outsider-authored free text when listing skills/versions through Agent Guard, because an external user can submit arbitrary search/query input (e.g., --name <PATTERN> / cursor paging data) that is then sent into and returned from the catalog/registry APIs and ingested/processed by the LLM as part of the result handling.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 1.00). The skill runs npx to fetch and execute the @jfrog/agent-guard package at runtime and specifies the npm registry URL https://releases.jfrog.io/artifactory/api/npm/coding-agents-npm/, which will be used to download and run remote code that controls catalog discovery/provisioning.