Навыки best-practices История аудитов
📦

История аудитов

best-practices - 1 аудит

Версия аудита 1 Последняя

22 окт. 2026 г., 13:23

All 95 static findings are false positives involving documentation examples, explicitly discouraged patterns, or legitimate audit references. No evidence found of prompt injection, credential theft, or malicious execution. Two semantic findings identify security guidance errors concerning Permissions-Policy delivery and structuredClone behavior.

2
Просканировано файлов
677
Проанализировано строк
7
Пункты проверки
0
Ложные срабатывания проигнорированы

Подтверждённые проблемы безопасности (2)

Средний
Permissions Policy Examples Do Not Enforce Restrictions
The skill recommends meta http-equiv="Permissions-Policy" to restrict geolocation, camera, and microphone. Browsers require the HTTP response header, so copied examples leave these policy restrictions unenforced.
Both examples use an unsupported meta delivery mechanism. This creates false assurance about feature restrictions, although browser permission prompts still apply.
Средний
Cloning Is Incorrectly Presented as Dangerous-Key Sanitization
The reference states that structuredClone drops __proto__ and functions, then labels its result deepSafe. Cloning preserves __proto__ data keys, so later unsafe merges can still expose prototype-pollution risks.
A local runtime check confirmed that structuredClone preserves an own __proto__ key and throws DataCloneError for functions. Cloning alone does not sanitize merge inputs.

Факторы риска

⚡ Содержит скрипты (7)
⚙️ Внешние команды (50)
🌐 Доступ к сети (18)
📁 Доступ к файловой системе (1)
🔑 Переменные окружения (2)
Аудитор:: codex