yichen-wechat-local-vault

Warn

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The script extract_keys.py performs dynamic code generation and execution by creating a temporary Python host script at runtime and executing it using the system interpreter to manage Frida instrumentation.
  • [DYNAMIC_EXECUTION]: extract_keys.py injects a JavaScript payload (FRIDA_JS) into the WeChat application process memory to hook the CCKeyDerivationPBKDF function. This is a form of runtime process manipulation used to capture derived encryption keys.
  • [EXTERNAL_DOWNLOADS]: During execution, extract_keys.py will attempt to install external Python packages (frida, frida-tools, pycryptodome) using pip if they are not already present on the system. This introduces a runtime dependency on external package registries.
  • [COMMAND_EXECUTION]: The skill frequently uses subprocess.run to execute shell commands, including codesign for modifying application signatures, pip for package management, and pgrep for process discovery.
  • [PRIVILEGE_ESCALATION]: extract_keys.py performs ad-hoc signing of the WeChat binary (codesign --force --deep --sign -). This technique is used to bypass macOS security protections, such as the Hardened Runtime, to enable process attachment and debugging on binaries that would otherwise be protected.
  • [INDIRECT_PROMPT_INJECTION]: The skill provides an attack surface for indirect prompt injection by ingesting and displaying chat logs, moments, and favorites which contain untrusted user-generated content. The developers have included a TRUST_MARKER (untrusted_snapshot_data) and a TRUST_WARNING in snapshot_reader.py and vault_cli.py, acknowledging that data processed from the vault should be treated as untrusted and could contain malicious instructions.
  • [DATA_EXFILTRATION]: The skill acts as a comprehensive extraction tool for sensitive personal data, including WeChat database encryption keys and the full history of user messages. While intended for local use, the ability to read and decrypt this volume of sensitive data creates a significant exfiltration surface for an autonomous agent.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 17, 2026, 03:57 AM