The Agent Skills Directory

[PROMPT_INJECTION]: The section titled 'AI FIRST Quality Principle' contains meta-instructions that attempt to override the internal behavior and operational procedures of the agent. It mandates a minimum of two complete iterations for all analysis and forbids 'single-pass output', using emphasizing language such as 'NO SHORTCUTS' and 'NEVER acceptable' to force the agent into a specific execution pattern contrary to standard efficiency guidelines.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process and analyze data from external sources, specifically through 'OSINT Collection' and automated integration of political data. This establishes a vulnerability surface where malicious instructions embedded in externally sourced text (such as politician statements or social media data) could influence the agent's behavior during the analysis phase.
Ingestion points: Data Foundation (OSINT Collection), SQL views (view_riksdagen_politician_summary), and political documents (document_element).
Boundary markers: The skill does not define or mandate the use of delimiters (e.g., XML tags or triple backticks) to separate untrusted external data from the agent's internal instructions.
Capability inventory: Use of SQL queries to query internal databases and Java-based analysis services for evaluating hypotheses.
Sanitization: No specific sanitization or filtering logic is prescribed for handling the external data before it is processed by the analytical techniques.
[COMMAND_EXECUTION]: The skill provides structured SQL templates for querying political databases (e.g., 'vote_data', 'sweden_political_party'). While these are intended for legitimate intelligence analysis, they demonstrate the agent's capability to generate and potentially execute data manipulation or extraction commands based on analytical needs.

intelligence-analysis-techniques