The Agent Skills Directory

[SAFE]: The skill is designed to perform data analysis and summarization of financial records. It includes specific instructions for the agent to verify the integrity of the data (e.g., checking if books are closed) and to avoid making assumptions about tax rates.
[INDIRECT_PROMPT_INJECTION]: The skill processes external financial data, which presents a theoretical attack surface for indirect prompt injection. However, the risk is minimal due to the following evidence chain:
- Ingestion points: Reads invoices, invoice_items, tax_rates, and journal_entries from the workspace.
- Boundary markers: Not explicitly defined in the instructions.
- Capability inventory: The skill utilizes data querying tools (well:querying-well-data) for analysis but lacks dangerous capabilities such as file system writing or shell command execution.
- Sanitization: Includes robust instructions to "surface uncertainty," "flag it for human review," and strictly "read the rate from the data" rather than guessing or hardcoding, which provides strong defensive logic against adversarial data.

vat-summary