social

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 the “listening” daily triage loop: it pulls public posts/comments from sources like Reddit/HN/Bluesky and browser-driven LinkedIn/X feeds, then parses their text into the agent context for scoring and drafting replies.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill's listening workflow explicitly runs runtime fetches (curl and dev-browser) against external endpoints — e.g. https://www.reddit.com/r/SaaS/new.json, https://hn.algolia.com/api/v1/search_by_date?... and https://public.api.bsky.app/xrpc/app.bsky.feed.searchPosts?... as well as LinkedIn/X feed URLs (e.g., https://linkedin.com/feed/hashtag/TOPIC/ or x.com/search?q=...) — and the skill injects that fetched content into the agent's context to score posts and draft replies, so external content directly controls agent outputs at runtime.