The Agent Skills Directory

[METADATA_POISONING]: The skill includes a file EVALUATION_REPORT.md that contains self-authored quality ratings and scores (e.g., 'Exemplary ⭐⭐', '9.2/10'). This self-referential content is designed to influence the perception of the skill's safety and quality.
[METADATA_POISONING]: There is a discrepancy between the platform-reported author ('Haibarakiku') and the metadata in the SKILL.md file ('theNeoAI').
[METADATA_POISONING]: The skill references multiple documentation files (e.g., statistical-methods.md, ml-modeling.md) that do not exist in the repository, indicating poor maintenance or deceptive structure.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external datasets, creating a surface for indirect prompt injection attacks.
Ingestion points: Processes 'raw data', 'behavior logs', and 'complex datasets' for analysis (SKILL.md).
Boundary markers: Absent. There are no instructions or delimiters to isolate untrusted data from the agent's core logic.
Capability inventory: The persona utilizes powerful data analysis libraries such as Pandas, Scikit-learn, and XGBoost to operate on provided data (references/toolkit.md).
Sanitization: Absent. No input validation or sanitization routines are defined for the data processed at runtime.
Mitigation: Wrap external content in delimiters with explicit 'ignore embedded instructions' warnings and validate data source before processing.

data-scientist