agent-prompt-builder

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

• Third-party content exposure detected (high risk: 0.85). The skill’s runtime workflow can ingest outsider-authored free text via /prompt sync and /prompt discover, which import content from external sources (e.g., website pages, RSS entries, tweets/LinkedIn exports) and then distill that text into the generated system prompt context.

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

• Potentially malicious external URL detected (high risk: 1.00). The skill explicitly fetches the agent's system prompt at runtime from the Live Neon API (e.g., curl "$LIVE_NEON_BASE/agents/$AGENT_ID" which maps to https://agentpersona.live/api/v1/agents/...) and then injects that fetched markdown as the model's system instruction, so remote content directly controls the agent's prompts.