The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest large amounts of untrusted data from external sources, creating a vulnerability surface for indirect prompt injection.
Ingestion points: In references/task1-company-research.md and references/task2-financial-modeling.md, the agent is instructed to fetch data from company websites, SEC EDGAR filings, earnings transcripts, and social media profiles (LinkedIn).
Boundary markers: The instructions do not specify the use of delimiters or explicit boundary markers when interpolating this external content into the report assembly phase.
Capability inventory: The skill has extensive file-writing capabilities (DOCX/XLSX skills) and can execute Python code for charting, as seen in references/task4-chart-generation.md and references/task5-report-assembly.md.
Sanitization: No instructions are provided for sanitizing or filtering instructions that might be embedded in the external financial documents.
[DYNAMIC_EXECUTION]: The workflow in references/task4-chart-generation.md involves the dynamic generation and execution of Python scripts to create financial visualizations.
The skill provides code templates to be populated with data extracted from the financial model and then executed to produce PNG files.
This pattern is a standard use of plotting libraries but constitutes dynamic execution of generated scripts.
[EXTERNAL_DOWNLOADS]: The skill instructs the agent to install several external Python packages in references/task4-chart-generation.md.
Packages include: matplotlib, seaborn, pandas, numpy, and plotly.
These are well-known, standard libraries used for data analysis and visualization.

initiating-coverage