The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/report_builder.py includes a routine to install the 'markdown' package via pip if it is not present in the environment. This is performed using an os.system() call to execute the installation command. Because the target is a well-known and reputable library from a standard registry, this is categorized as a legitimate use of dynamic dependency management.
[COMMAND_EXECUTION]: The skill's operational workflow requires the agent to execute specific local scripts for data processing and report packaging.
Evidence: Instructions in SKILL.md to run calculator/scripts/dasha_query.py for data retrieval and scripts/report_builder.py for final document generation.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection because it ingests and processes data from external files that could potentially contain untrusted content.
Ingestion points: Files structured_data.md and user_context.md are read during the analysis and report generation phases.
Boundary markers: The skill includes explicit instructions for "Blind Audit Principles" and "Logic Isolation" to ensure the agent maintains objectivity and disregards any biases or instructions embedded in the user's personal context files.
Capability inventory: The skill possesses capabilities to write files and execute shell commands through the provided utility scripts.
Sanitization: The skill relies on complex instructional guardrails rather than programmatic content sanitization to handle potentially malicious input in processed data files.

vedic-core