quickstart

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 1.00). The skill intentionally exfiltrates user data to a remote server (telemetry) and downloads+executes a remote bootstrap script, creating clear data-exfiltration and remote-code-execution/backdoor risks.

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

• Third-party content exposure detected (high risk: 0.85). Yes: the workflow fetches a personalized runbook from a remote URL at runtime (“Read it (WebFetch)”), and that fetched free-text content is then ingested into the agent’s LLM context, creating an indirect prompt-injection risk from outsider-authored web content.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill's bootstrap script curl-downloads a shell script from the Convex site and then runs it with bash (see "$BASE/quickstart-bootstrap" / https://basic-anteater-667.convex.site), which executes remote code at runtime.