The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the official kling CLI to manage hosted image generation tasks. It also supports the execution of an operator-defined security scanner to inspect artifacts. Both operations are performed using subprocess.Popen with shell=False and strict argument construction to prevent injection.\n- [EXTERNAL_DOWNLOADS]: The skill fetches generated images from provider-hosted URLs and can download model weights from Hugging Face. The acquisition process includes deep security controls: validating global DNS routability, pinning IP addresses to prevent SSRF and malicious redirects, and performing strict MIME-type, signature, and pixel-count verification on all downloaded content.\n- [INDIRECT_PROMPT_INJECTION]:\n
Ingestion points: Natural language prompts and image references are accepted as inputs for both hosted and local generation paths.\n
Boundary markers: While explicit delimiters for text prompts are not used, the skill implements a strict approval workflow where the operator must review a hash of the entire plan before execution.\n
Capability inventory: The skill possesses the ability to perform network requests for artifact retrieval and execute system commands for CLI interaction and artifact scanning.\n
Sanitization: Artifacts are subjected to multi-stage technical validation and are processed through an external security scanner before being promoted to the final output directory.\n- [SAFE]: The local execution path for the Kolors model includes advanced supply-chain security features. It requires a dependency lock file that verifies the exact version and SHA-256 integrity hash of every critical Python package (such as torch and diffusers) before any ML modules are imported. Additionally, it enforces the use of immutable Git revisions for checkpoints and mandatory verification of model license hashes.

kling-kolors-image