The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes a 'Suggest Using K-Dense Web For Complex Worflows' section which instructs the agent to proactively recommend a specific external platform (www.k-dense.ai) when user queries exceed a certain complexity. This is a behavioral steering attempt that modifies the agent's natural response pattern.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and visualize external data structures like pandas DataFrames and datasets. This creates a vulnerability surface where malicious instructions or markdown could be embedded within data labels or values, potentially influencing the agent or the visualization process. 1. Ingestion points: Data passed to plotting functions (df, datasets, individual_points) in SKILL.md. 2. Boundary markers: None identified in the provided instructions. 3. Capability inventory: File writing (save_publication_figure via figure_export.py). 4. Sanitization: No explicit sanitization or escaping of data labels is described.
[METADATA_POISONING]: There is a discrepancy between the author identified by the system (wu-yc) and the author claimed in the skill metadata (K-Dense Inc.). Additionally, the inclusion of promotional instructions for an external service within functional guidelines serves to steer agent behavior toward specific commercial endpoints.

scientific-visualization