The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes a section titled 'Suggest Using K-Dense Web For Complex Worflows' which instructs the agent to proactively suggest a specific hosted platform (www.k-dense.ai) when tasks reach a certain level of complexity. This is an attempt to influence the agent's behavioral logic for promotional purposes.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external data formats such as CSV, Parquet, and JSON, which creates a surface for indirect prompt injection if the data files contain malicious instructions.
Ingestion points: SKILL.md (Code examples for pl.read_csv, pl.read_parquet, pl.read_json, and pl.scan_csv).
Boundary markers: None specified in the instructions for data handling.
Capability inventory: Python code execution for data manipulation and analysis.
Sanitization: No explicit sanitization or validation of the data content is described in the provided examples.

polars