The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run and subprocess.call across multiple scripts, such as web_design_picker.py and run_factory.py, to orchestrate build steps and media optimization using tools like ffmpeg. These operations are sanitized via functions like validate_slug and validate_output_stem in scripts/_common.py, which ensure that user-influenced inputs are limited to safe, portable alphanumeric strings.
[DATA_EXPOSURE]: In scripts/_common.py, the skill implements an archive_secrets function that proactively identifies sensitive files (e.g., .env, SSH keys) and private key markers. The packaging logic in scripts/package_delivery.py uses this to ensure that no credentials or secrets are inadvertently included in the final delivery archives.
[INDIRECT_PROMPT_INJECTION]: The skill ingests user-supplied content to build website concepts. While this creates a data ingestion surface, the output is restricted to static HTML files. The skill does not process this untrusted data in a way that could influence the AI agent's core behavior or instructions.
[SAFE]: The skill provides robust protection against path traversal and symlink attacks through functions like reject_symlinks and safe_archive_path. These checks ensure that file operations remain strictly within the project's root directory during building and packaging.

web-design-picker