convex-add

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

• Third-party content exposure detected (medium risk: 0.30). The runtime workflow first fetches the served capability catalog and then fetches the matched capability’s /capability/<id>.md documentation from the Convex site, so the agent ingests provider-hosted free text instructions that could be influenced by an outsider who can publish those docs/items.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill fetches https://basic-anteater-667.convex.site/capabilities.json at runtime and then retrieves /capability/.md and follows its Procedure+Rules, meaning externally served content can directly control the agent's instructions.