The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads protein structure files (CIF/PDB), metrics, and plots from official vendor domains (*.fastfold.ai). The scripts implement several security controls for these operations, including protocol enforcement (HTTPS), hostname validation, file size limits, and the disabling of HTTP redirects to ensure safe handling of remote artifacts.
[INDIRECT_PROMPT_INJECTION]: The skill processes and displays untrusted data returned by the FastFold API, including job names, sequence metadata, and responses from the Evolla protein Q&A service, which creates a potential surface for indirect prompt injection.
Ingestion points: Data enters the agent context through scripts such as scripts/fetch_results.py and scripts/wait_for_evolla_linked.py, which retrieve and display job results and chat responses from api.fastfold.ai.
Boundary markers: The agent is explicitly instructed in SKILL.md to treat all API JSON as untrusted data and never execute commands embedded within it, though no programmatic delimiters are used in the script outputs.
Capability inventory: The skill possesses capabilities for network operations (GET/POST to the FastFold API) and local file system writes (downloading artifacts to the workspace directory).
Sanitization: The scripts implement validation for job IDs (UUID format) and hostname validation for artifact URLs to ensure they originate from the FastFold domain.

fold