deepseek-websearch
Pass
Audited by Gen Agent Trust Hub on Jul 3, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes a local Python script (
scripts/deepseek_search.py) via the shell to process user search queries. The workflow described inreferences/core_workflow.mdshows user-provided keywords being passed as command-line arguments to this script. - [DATA_EXFILTRATION]: User queries are transmitted to an external API endpoint (
https://redfox.hk/story/api/deepSearch/) to perform the search. This is the core functionality of the skill and targets the vendor's domain. - [EXTERNAL_DOWNLOADS]: The skill requires the installation of the
requestslibrary viapip3to facilitate network communication. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from the web via search results, which could contain malicious instructions.
- Ingestion points: Data enters the agent context through the stdout of
scripts/deepseek_search.py, which fetches results from the RedFox API. - Boundary markers: No explicit delimiters are used in the provided instructions to separate search content from system instructions.
- Capability inventory: The skill uses subprocess calls to execute its search logic.
- Sanitization: The script returns raw JSON data from the API without additional filtering or sanitization of the search content.
Audit Metadata