monte-carlo-validation-notebook

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

• Third-party content exposure detected (high risk: 1.00). This skill fetches and parses arbitrary public GitHub PR content (see Phase 1: "gh pr view ... --json files" and "gh api repos///contents/<file_path>?ref=<head_sha>") and the mc-bridge README via the GitHub API, and it directly reads and interprets those untrusted, user-generated files to drive schema resolution, diff analysis, and which validation queries to generate, so third-party content can materially change agent behavior.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill runs gh API calls at runtime to fetch repo files (e.g. gh api repos///contents/<file_path>?ref=<head_sha>) which retrieve SQL and dbt_project.yml content that the skill uses directly to construct notebook queries and thus controls the agent's outputs.