The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes biomolecular sequences in FASTA format, which represents an ingestion surface for untrusted data that could potentially harbor malicious instructions in record headers.
Ingestion points: Single typed FASTA files containing protein, ligand, RNA, or DNA records as described in SKILL.md.
Boundary markers: None specified to differentiate data from instructions within the processed FASTA records.
Capability inventory: The skill utilizes Bash(python:*), Bash(uv:*), Read, and Write tools to execute folding models and manage files.
Sanitization: There is no mention of sanitizing or validating FASTA headers before they are processed by the underlying tools.
[EXTERNAL_DOWNLOADS]: The skill facilitates the download of external model weights and software packages required for Chai-1.
Evidence: SKILL.md and references/chai_usage.md specify that the chai_lab package is installed via uv and several gigabytes of model weights are downloaded from the official repository and cached locally on the first execution.
[COMMAND_EXECUTION]: The skill uses shell commands to invoke the prediction model and manage the environment.
Evidence: The SKILL.md frontmatter allows Bash access, and the instructions describe using uv run python and the chai-lab fold CLI for performing complex predictions.

alterlab-chai