The Agent Skills Directory

[PROMPT_INJECTION]: The skill's instructions are clearly defined and focused on data analysis tasks. There are no attempts to override system safeguards, bypass ethical guidelines, or extract internal system prompts.
[CREDENTIALS_UNSAFE]: The skill requires a REDFOX_API_KEY, which is handled securely. The documentation instructs users to provide this through environment variables or local configuration files (~/.openclaw/openclaw.json) rather than hardcoding it, which is the recommended practice for API key management.
[DATA_EXFILTRATION]: Network activity is restricted to the vendor's domain (redfox.hk) for the purpose of retrieving comment data and providing API documentation. No sensitive local data or user credentials are transmitted to external servers.
[REMOTE_CODE_EXECUTION]: The Python scripts (douyin_comment_search.py and backfill_html.py) use standard Python libraries like urllib.request and json. They do not download or execute remote scripts, and they do not use unsafe functions like eval() or exec() on external data.
[COMMAND_EXECUTION]: The skill executes local scripts and the system open command to process data and display the generated HTML reports. These actions are transparently documented in the workflow and serve the skill's stated purpose.
[INDIRECT_PROMPT_INJECTION]: The skill processes external comments from the Douyin platform. While ingestion of untrusted data is a potential vector for indirect prompt injection, the skill incorporates HTML escaping (escape_html) in the report generation script to mitigate cross-site scripting risks, and the analysis is focused on statistical and sentiment summarization which is typical for this category of tool.

douyin-comment