venice-characters

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

• Third-party content exposure detected (medium risk: 0.30). SKILL.md describes that the workflow fetches Venice public character catalog/reviews via authenticated endpoints (GET /characters, GET /characters/{slug}, GET /characters/{slug}/reviews), which include outsider-authored free text like public review messages that the agent/LLM can ingest at runtime.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill makes runtime calls to Venice API endpoints (e.g. https://api.venice.ai/api/v1/characters?search=philosopher&sortBy=highestRating&limit=20 and https://api.venice.ai/api/v1/characters/alan-watts), and those responses include the character system prompts that Venice injects into chat—i.e., remote content fetched at runtime directly controls the agent's prompts.