social-scraping

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill instructs an agent to perform a trigger-then-poll flow where a JWT token returned by the service must be embedded verbatim in subsequent fetch commands/URLs (e.g., ?token=...), which requires the LLM to output secret token values directly.

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

• Third-party content exposure detected (high risk: 0.90). The SKILL.md clearly instructs the agent to fetch and ingest user-generated content from public social platforms via StableSocial endpoints (e.g., https://stablesocial.dev/api/instagram/profile, /reddit/search, TikTok/Facebook/LinkedIn endpoints) and to analyze/synthesize that content as part of workflows, exposing the agent to untrusted third‑party data that could carry indirect prompt injections.