baoyu-wechat-summary
Warn
Audited by Gen Agent Trust Hub on Nov 3, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPRIVILEGE_ESCALATIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes the
wxbinary with various subcommands (history,contacts,sessions,members). It uses user-supplied group names in these commands; while it specifies sanitization for folder naming, it does not explicitly define sanitization for the shell command arguments themselves, which could lead to command injection if malicious names are provided. - [PRIVILEGE_ESCALATION]: The skill explicitly instructs the agent to run with
dangerouslyDisableSandbox: trueto bypass default security restrictions. It also documents troubleshooting steps involvingsudo, although it correctly instructs the agent not to execute these automatically. - [DATA_EXFILTRATION]: The skill is designed to access and read highly sensitive private communication data located in
~/Library/Containers/com.tencent.xinWeChat/and configuration data in~/.wx-cli/. - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted chat messages from WeChat, creating a surface for indirect prompt injection attacks where malicious messages could attempt to override agent behavior during the summarization or profile-updating phases. • Ingestion points: WeChat message history is retrieved via
wx historyand stored in$TMPDIR/wx-messages.jsonbefore being read by the agent. • Boundary markers: The skill uses a structured multi-round workflow and specific digest sections, but does not implement robust delimiters or 'ignore' instructions for the raw message content within the prompt. • Capability inventory: The agent has the ability to execute shell commands (wx) and write multiple file types (.md, .json, .jsonl) to the local filesystem. • Sanitization: The skill includes '防注入' (anti-injection) logic inreferences/group-memory.mdspecifically to filter out messages that attempt to give instructions to the bot, though the risk remains for complex or novel injection patterns.
Audit Metadata