cafci-fondos-comunes-argentina

Fail

Audited by Gen Agent Trust Hub on Sep 30, 2026

Risk Level: HIGHEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill fetches financial data in JSON and XLSX formats from official Argentine domains estadisticas.cafci.org.ar and api.pub.cafci.org.ar. It also utilizes defuddle.md as a proxy service to convert remote HTML content into Markdown.
  • [COMMAND_EXECUTION]: The instructions utilize standard system tools including curl, jq, and python3 for data retrieval and format conversion.
  • [DYNAMIC_EXECUTION]: The skill uses a Python one-liner (python3 -c) to parse HTML content. Although flagged by automated scanners due to the curl | python3 pattern, the executed code is a static script defined within the skill itself and does not execute instructions from the remote source.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data which could contain malicious instructions.
  • Ingestion points: Markdown converted by defuddle.md and HTML attributes from cafci.org.ar (Operation 5b and 6 in SKILL.md).
  • Boundary markers: Absent; external content is interpolated without explicit delimiters or warnings to the agent.
  • Capability inventory: The agent has access to shell commands (curl, jq, python3) and can write temporary files to /tmp (Setup section in SKILL.md).
  • Sanitization: Data is structured via jq filters and specific regex patterns in the Python parsing script, providing limited sanitization against command injection but not against natural language instruction injection.
Recommendations
  • HIGH: Downloads and executes remote code from: https://estadisticas.cafci.org.ar/fondos/${FONDO_ID}?clase=${CLASE_ID} - DO NOT USE without thorough review
Audit Metadata
Risk Level
HIGH
Analyzed
Sep 30, 2026, 02:01 PM