The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The module scripts/utils/script_utils.py contains an ensure_yaml_installed function that executes pip install pyyaml via subprocess.check_call if the package is not found at runtime. This allows for dynamic code installation from external registries.
[COMMAND_EXECUTION]: Orchestration scripts such as scripts/management/refresh_index.py and scripts/management/rebuild_index.py utilize subprocess.run() to execute other Python scripts within the skill's own directory tree to perform indexing and maintenance tasks.
[EXTERNAL_DOWNLOADS]: The scraper implementation in scripts/core/scrape_docs.py fetches documentation from external sources, primarily https://cursor.com/llms.txt, using the requests library to update local content.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface. Content is ingested from the web via scrape_docs.py and stored without sanitization. This content is then parsed by extract_metadata.py and used as context for the agent, potentially allowing malicious instructions embedded in remote documentation to influence agent behavior.
[SAFE]: The skill implements a robust security control in scripts/utils/script_utils.py. The resolve_base_dir function includes path traversal protection that validates all resolved directories are contained within the trusted skill directory or the project root.

cursor-docs