omg

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

• Third-party content exposure detected (high risk: 0.80). The skill explicitly ingests user-generated, untrusted annotation and review data as part of its runtime flow — e.g., polling http://localhost:4747/pending and processing agentation annotations (acknowledge → navigate via annotation.elementPath → apply fix) and reading plannotator feedback (/tmp/plannotator_feedback.txt) to gate PLAN/EXECUTE — so external user content is read and can directly change agent actions.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill's installer (scripts/install.sh) includes a direct curl|bash install: "curl -fsSL https://plannotator.ai/install.sh | bash", which would fetch and execute remote code at runtime and plannotator is a required dependency used to control the PLAN approval flow, so this is a high-confidence runtime remote-code risk (https://plannotator.ai/install.sh).