paw-mkt-sostac

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

• Third-party content exposure detected (high risk: 1.00). The skill's required auto-discovery workflow (references/auto-discovery.md and SKILL.md Phase Flow Architecture) explicitly runs agent-browser/WebFetch on public sites (e.g., Google SERP, Reddit, Quora, G2, Trustpilot, Meta Ad Library, competitor homepages, Wayback Machine) and directs the agent to extract and use that untrusted, user-generated third‑party content to form research findings that directly drive recommendations, tactic selection, and next actions—creating a clear channel for indirect prompt injection.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill's activation instructs installing and running agent-browser with "npm install -g agent-browser && npx playwright install chromium", which fetches and executes remote packages from the npm/playwright registries (e.g. https://www.npmjs.com/package/agent-browser) at runtime, making this a runtime external dependency that executes remote code.