The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data files in multiple formats (CSV, Excel, JSON, XML, YAML, TOML) via a wide array of specialized scripts.
Ingestion points: Data is loaded using pandas and other standard libraries across all scripts, notably in data_storyteller.py, survey_analyzer.py, and data_converter.py.
Boundary markers: Although the SKILL.md file provides operational guardrails for the agent, the scripts do not implement explicit prompt delimiters or specific instructions for the agent to disregard potential natural language commands embedded within the data rows.
Capability inventory: The skill possesses significant data processing capabilities, including generating narrative findings and writing multiple report formats (HTML, PDF, CSV). It lacks network access or arbitrary system command execution, which limits the potential impact of an injection.
Sanitization: The skill largely relies on structured DataFrames. While dataset_comparer.py uses html.escape for report generation, the main attack surface remains the agent's interpretation of untrusted text content during the storytelling process.

data-storyteller