The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests protein sequences from users or external sources to be processed by computational tools and submitted to the Adaptyv API, representing an indirect injection surface.
Ingestion points: FASTA-formatted sequences are provided in SKILL.md and reference/examples.md for processing.
Boundary markers: The FASTA format (using >header and sequence blocks) provides structural separation, but there are no explicit instructions for the AI to ignore instructions potentially embedded in sequence headers.
Capability inventory: The skill possesses network capabilities (using requests) and file-write capabilities (described in reference/examples.md for saving results).
Sanitization: The validate_fasta function in reference/examples.md performs basic sanitization by ensuring sequences only contain valid amino acid codes (ACDEFGHIKLMNPQRSTVWY).
[EXTERNAL_DOWNLOADS]: The skill documentation instructs the installation of several external Python packages for data processing and bioinformatics, including requests, python-dotenv, pandas, biopython, soluprot, soluble-mpnn, fair-esm, torch, colabfold, and numpy. It also makes network requests to the Adaptyv API hosted on AWS Lambda (kq5jp7qj7wdqklhsxmovkzn4l40obksv.lambda-url.eu-central-1.on.aws) and the NetSolP service hosted by the Technical University of Denmark (services.healthtech.dtu.dk).
[COMMAND_EXECUTION]: The skill provides code examples in reference/examples.md that execute file system operations, such as creating directories using os.makedirs() and writing experiment results to local JSON and CSV files using Python's open() and pandas.to_csv() methods.

adaptyv