gzh-subscribe

Pass

Audited by Gen Agent Trust Hub on Oct 22, 2026

Risk Level: SAFEPERSISTENCECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [PERSISTENCE]: The skill implements a scheduled task feature using launchctl on macOS (via a generated .plist file in ~/Library/LaunchAgents) and crontab on Linux/Windows. This is an intended and documented feature used to automatically fetch articles at 06:00 daily.
  • [COMMAND_EXECUTION]: The Python script assets/subscribe.py executes several shell commands using subprocess.run to manage scheduling and open reports:
  • launchctl load/unload for macOS task management.
  • crontab commands for Linux task management.
  • open (macOS) and xdg-open (Linux) to display the generated HTML report in the default browser.
  • [EXTERNAL_DOWNLOADS]: The skill documentation in references/alipay-recharge.md instructs the agent to install the @alipay/agent-payment package using npx if the payment bot is missing. This is a legitimate integration with a well-known service (Alipay) for handling API credit recharge.
  • [INDIRECT_PROMPT_INJECTION]: The skill fetches external article data (titles, summaries) from the redfox.hk API and interpolates it into an HTML template using simple string replacement. This creates a potential surface for local Cross-Site Scripting (XSS) if the API returns malicious content, though the risk is limited to the local environment where the report is opened.
  • [DATA_EXFILTRATION]: The skill communicates with the vendor's official API (https://redfox.hk/story/api/gzh/data/queryWorkList) to fetch article data. It also accesses a local configuration file (~/.qoder/apis/redfox.json) to retrieve the user's API key, which is standard behavior for a service-integrated skill.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 22, 2026, 11:38 PM