The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection attacks as it is designed to ingest and process data from external, untrusted sources.
Ingestion points: The agent is instructed to read and analyze user-supplied Excel models (referencing references/core/model-review.md) and to search for and ingest information from public financial filings and market data (specified in SKILL.md).
Boundary markers: Absent. The instructions do not provide specific delimiters or warnings to the LLM to ignore potentially malicious instructions embedded within the financial data or user models.
Capability inventory: The agent has the ability to write files to the local filesystem using openpyxl and to execute a local Python auditing script (scripts/check_workbook.py) via subprocess, providing a path for potential exploitation if the agent is manipulated.
Sanitization: Absent. There is no evidence of logic designed to filter or sanitize natural language instructions that might be present in the ingested data.
[COMMAND_EXECUTION]: The skill requires the agent to run a local auditing script, scripts/check_workbook.py, on generated workbooks using a Python interpreter.
[EXTERNAL_DOWNLOADS]: The skill instructions in SKILL.md direct the agent to install standard, well-known Python libraries (openpyxl and formulas) from public registries. These are recognized as safe, established tools for spreadsheet manipulation.
[DYNAMIC_EXECUTION]: The agent is tasked with dynamically generating Python code using the openpyxl library to produce .xlsx workbooks with live formulas. This is a low-risk behavior as it uses a standard library for a specific, intended purpose.

moolaski-financial-analyst