The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches data from external protein and glycan databases (GlyConnect, GlyTouCan) which is then incorporated into the agent's context. The absence of explicit boundary markers or content sanitization for this external data represents an attack surface for indirect prompt injection.
Ingestion points: API calls in SKILL.md (query_glyconnect) and references/glycan_databases.md (get_glycoprotein_info, get_glycan_compositions).
Boundary markers: None present in the instructions.
Capability inventory: The skill possesses shell execution capabilities (Bash) and Python script execution.
Sanitization: The skill does not implement validation or escaping for the retrieved JSON data.
[EXTERNAL_DOWNLOADS]: The skill interacts with well-known bioinformatics and research services, including the Max Planck Society (mpg.de), Expasy (expasy.org), and DTU Health Tech (dtu.dk). It also facilitates the installation of the glycoshield package via pip for protein simulation analysis.
[COMMAND_EXECUTION]: The skill provides command-line instructions for installing and executing the glycoshield tool to calculate glycan surface shielding, which is consistent with its stated purpose in glycoengineering.

alterlab-glycoengineering