The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The script scripts/bootstrap.py is responsible for fetching the designcraft-cli binary from GitHub Releases. The download logic strictly validates the source URL against the author's official repositories and ensures the use of HTTPS. The downloaded archive is subjected to a file size limit check before processing.
[REMOTE_CODE_EXECUTION]: The skill installs and runs a native binary on the host machine. To ensure the integrity of the remote code, the skill verifies the SHA-256 checksums of both the downloaded archive and the extracted executable against pinned hashes located in scripts/runtime.lock.json. Additionally, the ZIP extraction process explicitly checks for and rejects unsafe patterns like absolute paths, parent directory references ('..'), and symbolic links.
[COMMAND_EXECUTION]: The Python infrastructure (e.g., cli.py, bootstrap.py) interacts with the system using subprocess.run. All commands are executed with list-based arguments rather than shell strings, which mitigates the risk of command injection. A subcommand whitelist further restricts the operations the agent can perform via the CLI.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided design projects (.designcraft files) which represent a potential attack surface for indirect prompt injection. The skill addresses this by utilizing a rigid multi-step workflow that requires explicit object discovery and validates all proposed command parameters against a structured schema before they are passed to the native execution engine.
[DYNAMIC_EXECUTION]: The skill uses importlib in scripts/cli.py and scripts/commands.py to dynamically load internal modules. This is used for modularizing the bootstrap and gateway logic and relies on static paths within the skill's own directory, which is considered a safe and standard implementation for this type of integration.

designcraft-cli-layout