The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes transaction descriptions and financial data from external banking sources, which constitute untrusted input and create a potential surface for indirect prompt injection.
Ingestion points: Bank transactions are fetched via MCP tools or imported from user-provided CSV/Excel files as outlined in SKILL.md.
Boundary markers: The instructions explicitly require the agent to "review it before presenting it as fact" and to "invite corrections rather than asserting it is their spending," which serves as a human-in-the-loop validation step.
Capability inventory: The skill utilizes a local Python script (scripts/categorize_transactions.py) for data processing and provides instructions for generating CSV exports for tax purposes.
Sanitization: The documentation includes specific requirements for file encoding (UTF-8 with BOM) to ensure the integrity of Hebrew text in exported data.
[EXTERNAL_DOWNLOADS]: The skill documentation references the israeli-bank-scrapers library on the npm registry and multiple GitHub repositories for MCP server implementations, including eshaham/israeli-bank-scrapers, mottibec/israeli-bank-mcp, and glekner/il-bank-mcp. These references target established community projects and official package registries relevant to the skill's financial domain.
[COMMAND_EXECUTION]: The instructions include standard shell commands for setting up the execution environment, such as npm install, npm run build, and docker compose. These commands are intended for the local deployment of the necessary MCP components and are documented with clear operational context.

israeli-bank-connector