The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process data from the LabArchives Electronic Lab Notebook (ELN) and Inventory systems, creating a surface for indirect prompt injection if malicious instructions are embedded in notebook entries or files.
Ingestion points: notebook_operations.py (local container inspection) and remote ELN/Inventory API responses containing user-generated scientific data.
Boundary markers: The instructions in SKILL.md and references/api_reference.md explicitly direct the agent to treat XML, JSON, filenames, and notebook content as untrusted data and to ignore instructions found within them.
Capability inventory: The bundled scripts are network-free and contain no command execution or dynamic code evaluation features. The agent's capabilities are restricted by the skill's instructions.
Sanitization: notebook_operations.py includes robust sanitization for archive inspection, including member path validation (rejecting traversal and absolute paths), size limits, and basic XML protection (rejecting DTD and entity declarations).
[EXTERNAL_DOWNLOADS]: The skill references and fetches metadata from the official arXiv repository and API for citation purposes.
Fetches citation information from arxiv.org and its official export API.
This uses a well-known scientific service and does not involve the execution of remote code.

labarchive-integration