The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill uses dynamic path manipulation in multiple scripts (e.g., build_prashna_data.py, judge_prashna_standard.py) to modify sys.path. This allows it to import a core planetary engine from a sibling directory (vedic-calculator/scripts). This is a legitimate code-sharing pattern for a modular suite of skills from the same author.
[INDIRECT_PROMPT_INJECTION]: The skill represents an attack surface for indirect injection as it processes natural language questions from users about astrological events. However, it implements a secure multi-stage pipeline:
Ingestion: User questions enter through the question argument in Phase 1 (SKILL.md).
Boundary Markers: The logic is constrained by strict JSON schemas (prashna_plan.json) and specific output templates described in judgment-writing.md.
Capability Inventory: The skill executes local Python scripts to perform mathematical calculations and rule-matching. It does not perform network operations or access system credentials.
Sanitization: The LLM first acts as a semantic parser to extract structured data into a validated plan before any interpretation logic is executed. This prevents direct user input from influencing code execution or bypassing safety guidelines.

vedic-prashna