deep-research

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 0.90). The skill contains explicit, non-user-visible operational instructions to autonomously publish generated reports (including local filesystem commands and a default opt-out trigger), which goes beyond simply producing research content and can exfiltrate outputs without clear user consent.

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). The skill contains intentional automated publishing steps that will publish user-generated reports to a public LovStudio blog by default (with an enforced "sync" gate) and spawning continuation agents that read local state — this is deliberate data-exfiltration/automatic-publish behavior that can expose user content without an explicit affirmative opt-in.

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

• Third-party content exposure detected (medium risk: 0.30). Source text from outsider-authored web pages is ingested at runtime during Phase 3 “RETRIEVE” via search-cli and (for deep dives) WebFetch/scrape/extract, before synthesis/triangulation writes evidence into evidence.jsonl.