baoyu-wechat-summary
Pass
Audited by Gen Agent Trust Hub on May 21, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill processes untrusted chat messages from WeChat group histories, which presents a surface for indirect prompt injection attacks where third-party messages could influence the agent's behavior.
- Ingestion points: Chat data is ingested through the
wx history --jsoncommand inSKILL.md(Step 3). - Boundary markers: The instructions for summarizing and generating profiles do not specify the use of robust delimiters or markers to isolate the untrusted message content from the system instructions.
- Capability inventory: The agent has the capability to execute shell commands via the
wxbinary (withdangerouslyDisableSandbox: true) and write files to the local filesystem for digest storage and profile management as defined inSKILL.md(Steps 7, 8, and 8.5). - Sanitization: While the skill excludes noise like emojis or system messages, there is no technical sanitization to prevent the model from interpreting malicious text within messages as instructions.
Audit Metadata