brave-search

Pass

Audited by Gen Agent Trust Hub on Mar 24, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill connects to the Brave Search API (api.search.brave.com) to retrieve web search data.\n- [COMMAND_EXECUTION]: Provides several Bash scripts (search.sh, search_ja.sh, search_news.sh) that utilize curl and jq for API interaction.\n- [DATA_EXFILTRATION]: Reads the Brave Search API key from the local configuration file ~/.clawdbot/brave-search-config.json and includes it in headers sent to the official API endpoint.\n- [PROMPT_INJECTION]: The skill processes untrusted content from web search results, which represents an indirect prompt injection surface.\n
  • Ingestion points: Search results (titles, snippets, URLs) from the api.search.brave.com endpoint.\n
  • Boundary markers: No explicit delimiters are used in the scripts to isolate search data from the rest of the agent's context.\n
  • Capability inventory: File system access for configuration reading and network access for API queries.\n
  • Sanitization: The skill uses jq to parse JSON fields, but does not sanitize the text content of the results for potential instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 24, 2026, 04:23 PM