The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads approximately 5 GB of model weights and embeddings (e.g., ESM2-3B) from the official Chai Discovery infrastructure during the first inference call. This is a required step for the model's operation.
[COMMAND_EXECUTION]: The skill provides instructions and code examples for executing shell commands via the chai-lab CLI and running Python scripts using the chai_lab library to perform inference and file system operations.
[DATA_EXFILTRATION]: When the --use-msa-server option is enabled, sequence data is sent to the public ColabFold MMseqs2 server (api.colabfold.com). This is a documented scientific feature used to improve prediction accuracy.
[PROMPT_INJECTION]: The skill processes biological sequence data from FASTA files, which represents a potential surface for indirect prompt injection if the output data is later used to influence agent logic.
Ingestion points: complex.fasta (user-provided input file)
Boundary markers: Absent; sequences and SMILES strings are processed directly
Capability inventory: Subprocess execution via CLI, GPU-accelerated inference, local file writes to output_dir, and network access to api.colabfold.com
Sanitization: None documented in the skill's instructions

chai1