The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The file config.py contains hardcoded production API keys for TAVILY_API_KEY and JINA_API_KEY, which are then used for network requests.
[COMMAND_EXECUTION]: In index.py, the search_agent_reach function uses subprocess.run to call mcporter. The user-provided keyword is interpolated directly into a JSON string passed as a command-line argument, which could lead to argument injection vulnerabilities depending on how the receiving tool parses the input.
[COMMAND_EXECUTION]: In index.py, the tavily_summarize function executes an external Node.js script located at a specific absolute path (/root/.openclaw/workspace/skills/tavily-search/scripts/search.mjs), which introduces a dependency on the host environment's file structure and other potentially untrusted skills.
[EXTERNAL_DOWNLOADS]: The skill fetches data from arbitrary external platforms via the r.jina.ai reader service. While used for legitimate search purposes, this allows the skill to retrieve content from any URL at runtime.
[PROMPT_INJECTION]: The skill is vulnerable to Indirect Prompt Injection.
Ingestion points: Search results (titles and abstracts) are fetched from external platforms via r.jina.ai in index.py.
Boundary markers: None. The content is directly concatenated into the summary prompt.
Capability inventory: The skill uses subprocess.run to execute commands and scripts.
Sanitization: There is no sanitization or escaping of the content retrieved from search results before it is interpolated into the prompt used for the AI summary.

cn-ai-search