The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs users to install a Python package from an unverified personal GitHub repository (https://github.com/mcmero/labarchives-py) using git clone and uv pip install .. This pattern involves executing unvetted code from an external source during installation and runtime.
[EXTERNAL_DOWNLOADS]: The installation process for the labarchives-py wrapper fetches code from an external repository not associated with a known trusted organization or the skill author's official infrastructure.
[CREDENTIALS_UNSAFE]: The setup workflow in scripts/setup_config.py prompts users to enter sensitive API credentials, including the Access Password and External Applications Password, which are then stored in a local config.yaml file. Storing plaintext credentials on disk, even with restricted file permissions, poses a risk of exposure.
[DATA_EXFILTRATION]: The skill is designed to retrieve and transmit sensitive research data (notebook backups, entries, and attachments) to a configurable API endpoint. While this is functional, the ability to specify an arbitrary api_url in the configuration could be abused to redirect data to an unauthorized server.
[COMMAND_EXECUTION]: The skill relies on local Python scripts (scripts/notebook_operations.py, scripts/entry_operations.py) that perform file system writes and network operations based on user-supplied command-line arguments and configuration settings.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it ingests and processes untrusted data from the LabArchives API (e.g., notebook names, entry content, and comments) without explicit sanitization or the use of secure boundary markers. An attacker could potentially embed malicious instructions in shared notebook content to influence the agent's behavior during data processing.

alterlab-labarchive