The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests birth chart data from external structured_data.md files. This data is parsed using regular expressions but lacks robust sanitization or delimiters to prevent embedded instructions from influencing the agent's behavior during report generation.
Ingestion points: structured_data.md (Person A and B's birth data) and synastry_data.md (intermediate calculated data).
Boundary markers: Not implemented in the prompt templates or markdown structures.
Capability inventory: Shell command execution for scripts (python scripts/...) and file writing capabilities in the reports/ directory via the build_synastry_data.py script.
Sanitization: Data is extracted via regex, providing structural validation but no sanitization of the content against prompt injection attacks.
[COMMAND_EXECUTION]: The skill's workflow involves executing local Python scripts via shell commands (python scripts/validate_synastry_data.py ...). These commands utilize placeholders for filenames and user-provided labels (e.g., --a <A名>). This pattern presents a command injection surface if the agent does not properly escape these inputs before execution, potentially allowing for arbitrary command execution.

vedic-synastry