wx-favorites-report
Fail
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: HIGHCOMMAND_EXECUTIONPRIVILEGE_ESCALATIONDATA_EXFILTRATIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PRIVILEGE_ESCALATION]: The skill facilitates bypassing macOS Hardened Runtime protections by instructing the user to re-sign the target application bundle (`codesign --force --deep --sign
- ~/Desktop/WeChat.app`). This disables the system-level security controls designed to prevent unauthorized code injection and memory inspection.
- [DYNAMIC_EXECUTION]: The skill uses Frida to perform process injection into a third-party application. It injects a JavaScript interceptor into
libcommonCrypto.dylibto hook theCCKeyDerivationPBKDFfunction, allowing it to capture derived encryption keys directly from the application's memory space. - [DATA_EXFILTRATION]: The skill accesses and exposes sensitive local database files. Specifically, it targets
favorite.dblocated within the application's container path at~/Library/Containers/com.tencent.xinWeChat/Data/Documents/xwechat_files/*/db_storage/favorite/favorite.db. This file contains private user data, including messages, links, and nicknames. - [COMMAND_EXECUTION]: The skill requires the execution of several sensitive CLI tools, including
codesignfor security control modification andfridafor runtime process manipulation. - [INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection by processing untrusted XML content extracted from the local WeChat database to generate a report.
- Ingestion points: The
parse_xml_contentfunction inparse_favorites.pyprocesses raw XML strings from thexmlBufcolumn of thefav_db_itemtable. - Boundary markers: None identified; data is processed directly from the database into a JSON format and then into an HTML report.
- Capability inventory: The skill has the capability to read local files, access SQLite databases, and write to the filesystem (
/tmp/data.json,/tmp/report.html). - Sanitization: The skill uses the standard
xml.etree.ElementTreefor parsing andjson.dumpsfor output, which provide basic structured data protection but do not account for malicious instructions embedded in the user-controlled database content.
Recommendations
- AI detected serious security threats
Audit Metadata