The Agent Skills Directory

[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it reads metadata and full-text content from the Zotero library, which could contain instructions from external sources.
Ingestion points: scripts/zotero_api.py reads data from the Zotero local API at localhost:23119.
Boundary markers: No explicit delimiters are used for the retrieved data.
Capability inventory: File writing (download_attachment), file reading (import_pdf), and internal application scripting (_debug_bridge) are available in scripts/zotero_api.py.
Sanitization: Key validation via regex and string escaping for file paths are implemented to protect script templates.
[COMMAND_EXECUTION]: The skill uses the Better BibTeX extension's debug-bridge to execute JavaScript within the Zotero application.
Evidence: The attach_file, create_collection, and delete_collection methods in scripts/zotero_api.py construct JavaScript strings and send them to the local Zotero server.
Protection: Malicious injection into these scripts is prevented by validating Zotero keys against a strict 8-character alphanumeric regex and properly escaping file paths.

zotero-local