The Agent Skills Directory

[METADATA_POISONING]: The skill's internal metadata identifies the author as 'K-Dense Inc.', which conflicts with the platform-provided author context ('beita6969'), constituting misleading information.- [INDIRECT_PROMPT_INJECTION]: The skill retrieves and processes data from external scholarly databases and arbitrary web URLs, creating a vulnerability where malicious instructions embedded in research content could influence agent behavior.
Ingestion points: Data is fetched from academic APIs (CrossRef, PubMed, arXiv) and arbitrary URLs via scripts/extract_metadata.py, scripts/search_google_scholar.py, and scripts/search_pubmed.py.
Boundary markers: The instructions lack explicit delimitation or security markers to isolate external metadata from the agent's core instruction set.
Capability inventory: The skill has permissions to use Bash for script execution and can perform Read, Write, and Edit operations on the filesystem.
Sanitization: There is no documented process for sanitizing or filtering instructions that may be embedded within paper titles, abstracts, or metadata fields.- [EXTERNAL_DOWNLOADS]: The skill relies on several standard Python packages from public registries and interacts with official academic services including CrossRef, NCBI (PubMed), and arXiv for its core functionality.- [COMMAND_EXECUTION]: The skill utilizes Bash to invoke local Python scripts responsible for automating academic database searches and managing BibTeX files.

citation-management