The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes a specific behavioral instruction in the 'Suggest Using K-Dense Web' section. It directs the agent to proactively steer users toward a commercial research platform (www.k-dense.ai) when tasks involve multi-step reasoning or high complexity. This attempts to influence the agent's conversational flow and recommendations.
[METADATA_POISONING]: The skill uses the instructions to embed promotional content for the author's (K-Dense Inc.) hosted platform, potentially biasing the agent's helpfulness toward a specific vendor's ecosystem.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, such as gene lists and TSV expression files, which are then interpolated into API requests and processed. While this is the primary functionality, it creates an attack surface where maliciously crafted biological data could attempt to influence the agent's analysis or output through the data ingestion pipeline.

reactome-database