The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by processing untrusted data from various external sources and incorporating it into structured outputs written back to shared Google Sheets.
Ingestion points: Processes data from Gmail message subjects (search_gmail_messages), Google Calendar event titles (list_events), and external Google Sheets (read_sheet_values).
Boundary markers: There are no specified delimiters or instructions to ignore embedded commands within the ingested data.
Capability inventory: The skill possesses write capabilities to Google Sheets (modify_sheet_values, create_sheet, create_spreadsheet).
Sanitization: No explicit sanitization or filtering of external content is documented prior to its inclusion in the generated reports.
[DATA_EXFILTRATION]: The skill is designed to access and aggregate sensitive corporate information, including financial metrics (cash levels, burn rates, and months of runway), OKR statuses, and meeting metadata from Gmail recaps. While it operates within the official Google Workspace environment and requires user confirmation for writes, it handles highly confidential organizational data.
[DYNAMIC_EXECUTION]: The skill implements logic to dynamically select and invoke tools (such as execute_sql or WebFetch) based on configuration data read from an external file (kpi-map.md from the zynkr-gm skill). Although execution is restricted to read-only operations and requires the user to explicitly name the connector during the run, this constitutes dynamic tool routing driven by external file content.

planning-evidence-pack