Historial de auditorías

nano-banana-2 - 1 auditoría

1 nov 2026, 14:40

All 51 static findings are false positives covering Markdown, documented service calls, normal credential storage, and model descriptions. A separate semantic finding identifies unsafe prompt interpolation in shell examples, compounded by an unsupported assurance that prompt content cannot cause shell injection. The external CLI implementation was unavailable, so its claimed protections were not independently verified.

Archivos escaneados

207

Líneas analizadas

Elementos de revisión

Falsos positivos ignorados

Preocupaciones de seguridad confirmadas (1)

Alto

SKILL.md:76-79 SKILL.md:84-93 SKILL.md:203

Unsafe Shell Interpolation Guidance for User Prompts

Examples embed '<user prompt>' inside single-quoted JSON shell arguments without escaping guidance. Verbatim substitution permits apostrophes to terminate shell quoting and introduce commands. The claim 'No shell injection surface from prompt content' overlooks parsing by the calling shell.

The examples visibly place prompt placeholders inside shell single quotes, and the safety claim ignores caller-side parsing. Exploitation depends on verbatim substitution; no executable wrapper was supplied.

Factores de riesgo

Auditado por: codex

← Volver a la habilidad