install-bensz-skills

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

• Third-party content exposure detected (high risk: 0.90). The installer explicitly clones and reads remote GitHub repositories listed in config.yaml (see scripts/install.py::_download_remote_source and _compare_remote_skills which parse SKILL.md and compute MD5), and then uses those repo files to decide and perform installations (including an automatic --remote --auto mode), so untrusted third‑party repo content can be ingested and materially influence what skills are installed and thus agent behavior.

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

• Potentially malicious external URL detected (high risk: 0.90). The installer’s remote mode explicitly runs git clone at runtime to fetch configured repositories (e.g. https://github.com/huangwb8/skills, https://github.com/huangwb8/ChineseResearchLaTeX, https://github.com/anthropics/skills) and then copies those repos into system-level skill directories (~/.codex/skills, ~/.claude/skills), meaning remote content fetched during runtime can directly change agent skills/instructions and thus control prompts/behavior.