The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The scripts/bootstrap.py script downloads a native binary archive from GitHub releases. The download is restricted to official vendor repositories (storytold/designcraft or full-aigc-skills/designcraft-skills) and follows the platform's trusted source guidelines. Verification is enforced by checking the archive's SHA256 hash against a lockfile before any content is extracted.
[REMOTE_CODE_EXECUTION]: The skill installs and executes a remote binary (designcraft-cli). This risk is mitigated by strict version locking, checksum verification of the binary itself, and isolation of the runtime environment to a local user directory without requiring elevated privileges.
[COMMAND_EXECUTION]: Command execution is handled via subprocess.run using argument lists rather than shell strings. This design prevents shell injection vulnerabilities. The skill also enforces a whitelist of allowed subcommands and validates parameters against a structured schema.
[DYNAMIC_EXECUTION]: The entry point scripts cli.py and commands.py use importlib to load internal modules from the skill's local directory. This is used for modular code organization and relies on static, relative paths to its own distributed files.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data in the form of command plans and project files. It implements a 'strict_json' parser to prevent duplicate key or non-finite number exploits and validates all command inputs against a schema to ensure data conform to expected types and ranges.

designcraft-cli