The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The scripts/bootstrap.py module downloads the lightcraft-cli binary from GitHub. Security is maintained by restricting downloads to specific vendor organizations and verifying the SHA256 checksum of the archive before extraction.
[REMOTE_CODE_EXECUTION]: The skill installs and executes an external binary (lightcraft-cli). This is a core feature of the skill, and risks are mitigated through strict download validation, local installation without elevated privileges, and checksum verification of the executable binary.
[COMMAND_EXECUTION]: The skill uses the subprocess module in several scripts, including scripts/native_process.py, scripts/artifacts.py, and scripts/command_gateway.py, to run the CLI tool and the system utility sips. Command arguments are whitelisted or strictly validated before execution.
[DYNAMIC_EXECUTION]: Modules within the skill directory are loaded at runtime using importlib.util in scripts/cli.py and scripts/command_gateway.py. This facilitates modular code organization within the skill's own local scope.
[INDIRECT_PROMPT_INJECTION]: The skill processes external image files and library metadata, presenting an attack surface for indirect injection via untrusted data.
Ingestion points: scripts/artifacts.py ingests image data for decoding and metadata extraction; scripts/command_gateway.py processes library data.
Boundary markers: The skill uses structured JSON plans and execution receipts to maintain a clear separation between instructions and data.
Capability inventory: Subprocess execution is limited to the verified CLI binary and established system tools like sips.
Sanitization: Extensive validation is performed, including JSON schema checks for all command parameters and strict path resolution to ensure operations remain within the designated output root.

lightcraft-cli-export