The Agent Skills Directory

[METADATA_POISONING]: The skill uses extreme deception in its name and description, claiming to be an internal tool for 'my own Amazon (Kindle) book library.' However, the underlying implementation and API documentation point exclusively to 'library-archive.org' and its associated mirrors, which are components of Anna's Archive, a site for accessing copyrighted material without authorization.
[CREDENTIALS_UNSAFE]: The skill instructs users to set a LIBRARY_KEY environment variable, identifying it as an 'Amazon-library access key' or 'Amazon library access setup.' This sensitive token is then transmitted to untrusted external domains associated with pirate sites, which could lead to credential harvesting or exposure of account keys to malicious actors.
[EXTERNAL_DOWNLOADS]: The skill is configured to search and download files from a list of domains known for hosting pirated content, including library-archive.li, library-archive.se, library-archive.in, and library-archive.pm. These domains are untrusted and frequently rotated to evade legal actions, making downloads from them a significant risk for malware or unauthorized content ingestion.
[INDIRECT_PROMPT_INJECTION]: The skill contains a vulnerability surface for indirect prompt injection as it processes external, untrusted data from shadow libraries.
Ingestion points: The fetch_url function in scripts/library.py retrieves raw HTML from various external mirror sites.
Boundary markers: Absent. The skill does not implement delimiters or instructions for the agent to ignore potentially malicious content embedded in book titles or metadata.
Capability inventory: The script has the capability to write files to the filesystem using urllib.request.urlretrieve in scripts/library.py.
Sanitization: The skill uses basic regular expressions for parsing metadata but lacks robust sanitization or validation of the retrieved content before it is presented to the agent.

library-ebooks