The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external files and services, providing a surface for indirect prompt injection if the ingested data contains malicious instructions.\n
Ingestion points: The skill loads data using pandas.read_csv in csv.md, openpyxl.load_workbook in excel.md, and the Google Sheets API values.get method in google-sheets.md.\n
Boundary markers: The instructions lack explicit markers or warnings to the agent to distinguish between tabular data and executable instructions.\n
Capability inventory: The skill is capable of writing to the local filesystem and performing network operations via the Google Sheets API.\n
Sanitization: No data sanitization or validation routines are specified for the cell contents retrieved from external sources.\n- [COMMAND_EXECUTION]: The skill uses shell commands to initialize its required directory structure.\n
Evidence: Instructions in SKILL.md direct the creation of a local storage environment using mkdir -p ~/spreadsheet/{projects,templates,exports}.\n- [DATA_EXFILTRATION]: The skill performs network requests to synchronize data with the Google Sheets API.\n
Evidence: The skill communicates with the well-known service domain at https://www.googleapis.com/auth/spreadsheets for reading and writing spreadsheet values.\n- [CREDENTIALS_UNSAFE]: The skill manages authentication via local credential files, which involves accessing sensitive file paths.\n
Evidence: Instructions in google-sheets.md and memory-template.md reference credentials.json and ~/credentials.json for managing Google service account keys.

Spreadsheet