The Agent Skills Directory

[SAFE]: The skill implements legitimate election forecasting logic for the Swedish electoral system, including the Modified Sainte-Laguë seat allocation method and weighted polling aggregation. All operations are local or directed toward a provided database connection.
[INDIRECT_PROMPT_INJECTION]: The skill has an ingestion surface for external data as it reads polling figures, economic indicators, and government status from a database via SQL queries in SKILL.md.
Ingestion points: Python methods aggregate_polls_weighted, structural_forecast_model, and calculate_economic_vote query several database tables (opinion_polls, world_bank_data, current_government_status).
Boundary markers: Absent. The data is retrieved and processed as numeric values.
Capability inventory: The skill performs data analysis but lacks shell execution (subprocess), file-writing, or outbound network operations.
Sanitization: The skill processes data primarily as numeric types (floats/integers) for statistical modeling, which naturally mitigates the risk of instructions being interpreted by the agent during calculation.
[EXTERNAL_DOWNLOADS]: The skill references several external resources for documentation and data sourcing. These include official Swedish government websites (val.se, riksdagen.se), academic sources, and well-known polling organizations (novus.se, sifo.se). It also references the vendor's own security policy repository on GitHub and official GitHub workflow documentation. These references are documented neutrally as they point to trusted or well-known organizations and services.
[REMOTE_CODE_EXECUTION]: The skill utilizes standard Python libraries for data analysis (pandas, numpy, scipy, scikit-learn). No unpinned or suspicious third-party packages are requested, and there is no pattern of downloading and executing scripts from remote sources.

electoral-analysis