The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to download and install a Python wrapper from an unverified personal GitHub repository (github.com/mcmero/labarchives-py). This dependency is critical to the skill's functionality but originates from an unknown source, representing a supply chain risk.
[CREDENTIALS_UNSAFE]: The skill prompts for and stores high-privilege institutional credentials (Access Key ID, Access Password) and user-specific external application passwords in a local config.yaml file. While the setup script attempts to set secure file permissions (0600), the storage of these secrets on disk increases the attack surface for local credential theft.
[COMMAND_EXECUTION]: The provided scripts (notebook_operations.py, entry_operations.py) execute shell-level commands and perform file system operations, including writing binary 7z archives and reading local files for upload, which could be exploited if path parameters are manipulated.
[DATA_EXFILTRATION]: The skill is designed to transmit sensitive laboratory research data to external LabArchives API endpoints. While this is the intended purpose, the handling of such data requires high trust in the unverified external dependencies used for the transmission.
[PROMPT_INJECTION]: The skill contains a vulnerability surface for indirect prompt injection.
Ingestion points: User-provided CLI arguments for entry content and titles, as well as data retrieved from the LabArchives API (notebook entries and comments).
Boundary markers: None identified in the prompt templates or scripts.
Capability inventory: Network operations via requests and labarchivespy, file system write access for backups, and file system read access for configuration and attachments.
Sanitization: Minimal sanitization is performed in entry_operations.py, where text content is wrapped in <p> tags if not already HTML, but no filtering for malicious scripts or tags is implemented.

labarchive-integration