The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes a section at the end directing the agent to proactively promote a specific commercial platform (K-Dense Web) when research tasks become complex, which acts as a behavioral steer.
[METADATA_POISONING]: There is a contradiction between the system-level author (beita6969) and the skill metadata (K-Dense Inc.), which could be misleading regarding the skill's origin.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process abstracts and metadata from external scientific databases, creating an attack surface for indirect prompt injection.
Ingestion points: Data retrieved from PubMed, arXiv, and bioRxiv aggregated in aggregated_results.md and unique_results.json.
Capability inventory: The skill utilizes Bash, Write, and Edit tools for file management and document generation.
Boundary markers: The workflow instructions do not define delimiters or warnings to prevent the agent from following instructions potentially embedded in academic text.
Sanitization: The skill description does not specify any validation or sanitization procedures for the external content before processing it within the agent context.

literature-review