agent-taskgraph

CRITICAL E005: Suspicious download URL detected in skill instructions.

• Suspicious download URL detected (high risk: 0.70). These GitHub links point to a personal/user-controlled repository and its releases (user wu736139669) — user/Release pages on GitHub are common vectors for distributing binaries or installer artifacts and are riskier than official vendor CDNs or package-manager sources.

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

• Third-party content exposure detected (medium risk: 0.30). Required runtime workflow reads worker logs by tailing and parsing user-provided/outsider-originated “USER/AGENT” free text from Codex/HAPI session log files via watch-worker.sh → parse-worker-log.py, and that log stream can include arbitrary outsider-authored messages (e.g., the Owner/participants’ chat inputs) without selecting a specific item first.