The Agent Skills Directory

[SAFE]: The skill utilizes well-known and reputable scientific libraries, including DECIMER and huggingface_hub, for image recognition tasks. Dependencies are clearly defined in requirements.txt and are sourced from standard registries.
[SAFE]: Data handling is restricted to reading user-specified image files and writing results to a designated output directory. Network communication is limited to the HuggingFace Inference API, which is a well-known service, for performing model inference on the provided images.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it processes untrusted image data that could potentially contain adversarial textual instructions.
Ingestion points: The scripts/mol_image_to_smiles.py script accepts user-provided file paths via the -i/--input command-line argument and reads the file content.
Boundary markers: No delimiters or explicit instructions to ignore embedded textual content are provided when the data is passed to the recognition models.
Capability inventory: The skill has the capability to read local files, write JSON results to the file system, and perform network requests to HuggingFace.
Sanitization: No sanitization, validation, or filtering of potential textual content within the input images is performed before processing.

mol-image-to-smiles