lucid

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

• Third-party content exposure detected (medium risk: 0.30). The runtime workflow’s outsider-ingest path is scripts/lookup_docs.py --fetch which uses a URL derived from an outsider-supplied query/--fetch target to download arbitrary text (via urllib.request.urlopen from the provided URL or candidates like https://r.jina.ai/...), which the agent would then read/print as fetched content.

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

• Potentially malicious external URL detected (high risk: 0.90). scripts/lookup_docs.py is invoked at runtime (see SKILL.md) and performs live fetches of official Markdown from https://lucid.adonisjs.com (and via the jina proxy https://r.jina.ai/) — those fetched pages (e.g., https://lucid.adonisjs.com/docs/models.md) are pulled into the agent flow and can directly control prompts/instructions.