The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads compressed Lightcraft CLI archives from GitHub releases.
Evidence: scripts/bootstrap.py implements a trusted_release_url filter that restricts downloads to the storytold/lightcraft and full-aigc-skills/lightcraft-skills organizations on github.com.
Verification: Downloads are validated against hardcoded SHA256 hashes defined in scripts/runtime.lock.json before any extraction or execution occurs.
[COMMAND_EXECUTION]: The skill executes the downloaded native binary and system utilities (like sips on macOS) to perform its core functions.
Evidence: scripts/native_process.py utilizes subprocess.Popen to manage the execution of the Lightcraft CLI. scripts/bootstrap.py uses subprocess.run for installation verification.
Protection: scripts/cli.py implements a whitelist (ALLOWED) for subcommands to prevent the execution of unauthorized binary functions.
[DYNAMIC_EXECUTION]: The skill installs and runs an executable binary at runtime.
Evidence: scripts/bootstrap.py extracts the CLI binary to a local runtime directory and verifies it by executing a version check.
[INDIRECT_PROMPT_INJECTION]: The skill processes external JSON-formatted plan and receipt files, which represents an attack surface for indirect injection if the input files are attacker-controlled.
Ingestion points: scripts/commands.py reads plan files from user-supplied paths, and scripts/batch_scope.py processes scope and change requests.
Capability inventory: The skill can execute commands via native_process.py and write files via command_gateway.py.
Sanitization: The skill employs strict_json parsing (rejecting duplicate fields and non-finite numbers) and performs comprehensive JSON schema validation before processing inputs.

lightcraft-cli-setup