The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads a native CLI binary from GitHub releases (github.com/storytold/designcraft) during the first run. This is a legitimate dependency for the skill's primary function.
The bootstrap.py script implements high-security download practices, including hardcoded SHA256 hash verification for both the zip archive and the extracted binary.
It performs strict path validation during zip extraction to prevent directory traversal (ZipSlip) attacks.
[COMMAND_EXECUTION]: The skill executes the downloaded native binary using the subprocess module.
Arguments are passed as a list (argv) rather than a shell string, mitigating shell injection risks.
Execution is limited to a predefined allowlist of subcommands (e.g., run, describe, version).
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests and processes external DesignCraft project files.
Ingestion points: scripts/command_gateway.py reads project data via the native CLI's document.inspect command.
Boundary markers: There are no explicit delimiters instructed for the agent to ignore instructions embedded within project metadata or text objects.
Capability inventory: The skill can execute native commands and write files (file.saveAs) through the CLI wrapper.
Sanitization: The skill mitigates risks through strict JSON schema validation of all command parameters and structural checks of the execution plan before any native call is made.

designcraft-cli-document