The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from Gmail and bank statement PDFs which may contain malicious instructions. \n
Ingestion points: Bank statement PDF processing in Step 1 and Gmail sweep for invoices/receipts in Step 2 of SKILL.md.\n
Boundary markers: No explicit prompt boundary markers or 'ignore instructions' delimiters are defined for the ingested text.\n
Capability inventory: The skill has permission to append to Google Sheets (Transactions tab), upload files to Google Drive, and send HTML emails via Gmail.\n
Sanitization: Data is processed into a strict 16-column schema and verified against a balance identity gate in Step 4, but the skill lacks specific sanitization for natural language instructions embedded in processed documents.\n- [EXTERNAL_DOWNLOADS]: The skill retrieves historical currency exchange rates from a well-known CDN to calculate ledger amounts.\n
Source: Fetches JSON data from the fawazahmed0 currency-api via the jsdelivr.net CDN.\n
Evidence: references/ledger-contract.md describes using the currency API via jsdelivr.\n- [COMMAND_EXECUTION]: Utilizes a local system utility to extract text from bank statement PDFs.\n
Evidence: Step 0 of SKILL.md specifies using pdftotext with the layout flag to parse transaction tables while maintaining column structure.

zynkr-accounting