The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data through DataFrames and remote dataset loading, which establishes a surface for indirect prompt injection. If an attacker provides data containing instructions in labels or content, the agent might inadvertently execute them when processing the visualization output.
Ingestion points: Data ingested via pandas.DataFrame fixtures and the sns.load_dataset() function as noted in references/detailed-guide.md.
Boundary markers: The instructions do not specify any delimiters or warnings to prevent the agent from interpreting data content as instructions.
Capability inventory: The skill utilizes matplotlib's savefig method in references/detailed-guide.md to write visualization files to the local file system.
Sanitization: There is no explicit logic described for sanitizing data inputs or visualization labels to filter out potential injection strings.
[EXTERNAL_DOWNLOADS]: The skill includes usage of sns.load_dataset(), a function that fetches sample data from the Seaborn project's official remote repositories.
Evidence: references/detailed-guide.md mentions that sns.load_dataset downloads remote examples when they are not cached locally.
[METADATA_POISONING]: The skill's metadata contains conflicting information regarding its authorship and risk level.
Evidence: The YAML frontmatter in SKILL.md identifies the author as K-Dense Inc., which does not align with the registered author context. Furthermore, it includes a risk: critical metadata field which may be intended to misrepresent the skill's risk profile or influence automated scanning systems.

seaborn