wecom-cs-automation

Warn

Audited by Gen Agent Trust Hub on Mar 9, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
  • [COMMAND_EXECUTION]: The install.sh script performs automated system configuration that requires elevated privileges via sudo, including updating package repositories, installing PostgreSQL, and compiling source code.
  • [EXTERNAL_DOWNLOADS]: Fetches and builds the pgvector extension from its official GitHub repository (github.com/pgvector/pgvector) during the installation process.
  • [PROMPT_INJECTION]: The skill has a surface for indirect prompt injection where malicious instructions embedded in user queries could influence the agent's behavior.
  • Ingestion points: workflows/answer_question.py (processes external user questions).
  • Boundary markers: Absent; user queries are directly interpolated into the LLM prompt without delimiters or warnings.
  • Capability inventory: Database access (psycopg2), LLM communication (openai), and messaging via WeChat Work and Telegram.
  • Sanitization: No sanitization or validation of user-provided content is implemented before it is passed to the LLM.
  • [CREDENTIALS_UNSAFE]: The installation script creates a .env file template for storing sensitive information, such as WECOM_AGENT_SECRET and LLM_API_KEY, in plain text on the local filesystem.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 9, 2026, 10:14 PM