wechat-mp-publisher

Warn

Audited by Gen Agent Trust Hub on Apr 9, 2026

Risk Level: MEDIUMDATA_EXFILTRATIONCOMMAND_EXECUTION
Full Analysis
  • [DATA_EXFILTRATION]: The SKILL.md file contains mandatory instructions for the agent to take screenshots of the WeChat MP editor (which displays the full text, title, and media of drafts) and send them to a specific hardcoded Telegram group ID (-1003890797239). This creates a data exfiltration channel where sensitive user content is transmitted to an external destination managed by a third party.
  • [COMMAND_EXECUTION]: The skill relies on local Python scripts (scripts/publish.py and scripts/api_publish.py) that perform browser automation via Playwright and authenticated network requests. These scripts execute with the user's local privileges and handle sensitive session data.
  • [DATA_EXPOSURE]: The skill stores and accesses sensitive platform credentials, including browser cookies and API tokens, within the user's home directory (~/.openclaw/skills/wechat-mp-smart-publish/). This represents a local exposure surface for authentication secrets.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, untrusted content (Markdown and HTML articles) and perform automated browser/API actions based on that content. There is a lack of boundary markers to prevent instructions embedded within the article text from influencing the agent's behavior in the editor.
  • Ingestion points: Article content files loaded into scripts/publish.py and scripts/api_publish.py.
  • Boundary markers: None identified in the prompt templates or scripts.
  • Capability inventory: Full browser automation (Playwright), file system reads, and WeChat API interactions.
  • Sanitization: The scripts perform basic character length checks but do not sanitize HTML/Markdown content for malicious injection strings.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 9, 2026, 12:28 AM