The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from user-supplied CSV files and remote datasets to prepare training inputs for bio-foundation models.
Ingestion points: Data is read from user-provided file paths via the --data_path argument and from a remote TSV file in scripts/prepare_ribonn.py.
Boundary markers: Instructions do not define specific delimiters or warnings to prevent the agent from following natural language instructions potentially embedded within the CSV data columns.
Capability inventory: The skill executes Python scripts for data processing (prepare_ribonn.py) and model training (src.runner), both of which perform file system operations.
Sanitization: The prepare_ribonn.py script performs sequence validation (ACGT character check) and basic type/finite-value checks on labels.
[EXTERNAL_DOWNLOADS]: The data preparation script fetches a human translation-efficiency dataset from a public GitHub repository for demonstration purposes. This download targets a well-known service and is pinned to a specific commit hash for integrity.
[COMMAND_EXECUTION]: The skill provides instructions for the agent to execute shell commands to run Python scripts for dataset preparation and fine-tuning tasks using the src.runner module.

codonfm-finetune