The Agent Skills Directory

[DYNAMIC_EXECUTION]: The Python scripts (including scripts/build_prashna_data.py, scripts/calc_moon_vedic.py, scripts/calc_optional_kp.py, and scripts/calc_optional_tajika.py) utilize sys.path.insert(0, ...) to dynamically alter the Python module resolution path at runtime. They attempt to load a core engine dependency from a computed relative path (../vedic-calculator/scripts). This dynamic loading from computed paths makes the skill's execution environment and dependency chain difficult to verify statically and could be exploited to load malicious code if the directory structure is compromised.
[INDIRECT_PROMPT_INJECTION]: The skill ingests free-text question strings from users and writes them directly into persistent markdown and JSON artifacts (such as structured_prashna.md and kp_judgment.md) without sanitization. Because the agent is instructed to read these artifacts when processing follow-up queries (Phase 7), this establishes a vulnerability to indirect prompt injection.
Ingestion points: User-supplied input for the 'question' collected during Phase 0.
Boundary markers: The ingested text is placed into markdown documents using standard formatting (like tables or code blocks) but lacks specific protective delimiters or instructions to ignore embedded commands.
Capability inventory: The skill can execute shell commands to run Python calculations and write data to the local file system.
Sanitization: The implementation does not perform validation, escaping, or sanitization of the user-provided question before it is written to output files.

vedic-prashna