The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill performs direct modification of the Zotero SQLite database located at D:\software\zotero\zotero.sqlite. Although it includes a step to back up the database before modification, direct SQL manipulation (modifying tags, itemTags, itemData, and collectionItems tables) of a third-party application's data is a high-risk operation that can lead to data corruption or unintended side effects.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data by extracting metadata (DOI, title, author) from local PDF files to generate RIS files and database entries.
Ingestion points: Reads content and metadata from local PDF files located in user-specified directories.
Boundary markers: None specified to prevent the agent from interpreting instructions embedded in PDF metadata.
Capability inventory: Includes file system write access (backups and RIS files), subprocess execution (Python helper), and direct database write access.
Sanitization: No explicit sanitization or validation of the extracted PDF content is described before it is interpolated into database commands or RIS files.
[COMMAND_EXECUTION]: The skill invokes a Python script (zotero.py) located within the plugin's internal directory to manage Zotero status checks, target selection, and RIS imports.
[DATA_EXFILTRATION]: The skill accesses highly sensitive user data, specifically the entire Zotero Desktop library database (zotero.sqlite) and local document collections. While no network exfiltration was detected in the provided code, the level of access to personal research data is significant.

p-pdf2zotero