data-bird

This skill appears to implement legitimate CSV/Excel analysis, but the SKILL.md explicitly tells the agent to use shell curl to download arbitrary URLs (including internal / localhost) and to scan workspace directories. That pattern enables SSRF/internal scanning and potential data exfiltration if an attacker supplies a malicious URL or if the agent is tricked into fetching internal endpoints. Before installing or enabling this skill: - Prefer a version that uses the platform's safe HTTP fetch helper (web_fetch) or vetted SDK rather than exec curl. Ask the author to remove the explicit 'use curl' instruction. - Do not provide sensitive URLs, database credentials, or paths unless you run the skill in an isolated environment. - Audit config.yaml and ensure enable_mysql is false unless you intentionally provide DB credentials; limit max_rows/max_file_mb to safe values. - Run the skill in a sandboxed agent (no access to production internal networks) or review/execute the code in an offline environment first. - If you must use it, inspect the code paths that handle downloads and file I/O (scripts/main.py, report_artifacts.py, data_agent.py) and consider enforcing allowlists for download domains and filesystem paths.